Nonlocal Integral Formulations of Plasticity and Damage:

Survey of Progress
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

Zdeněk P. Bažant1 and Milan Jirásek2

Abstract: Modeling of the evolution of distributed damage such as microcracking, void formation, and softening frictional slip neces-
sitates strain-softening constitutive models. The nonlocal continuum concept has emerged as an effective means for regularizing the
boundary value problems with strain softening, capturing the size effects and avoiding spurious localization that gives rise to pathological
mesh sensitivity in numerical computations. A great variety of nonlocal models have appeared during the last two decades. This paper
reviews the progress in the nonlocal models of integral type, and discusses their physical justifications, advantages, and numerical
applications.
DOI: 10.1061/共ASCE兲0733-9399共2002兲128:11共1119兲
CE Database keywords: Plasticity; Damage; Cracking; Localization.

Introduction but their interaction can take place only on the level of balance
equations, through the exchange of mass, momentum, energy, and
Historical Beginnings entropy.
In reality, however, no material is an ideal continuum. Both
Most standard constitutive models for the mechanical behavior of natural and man-made materials have a complicated internal
solids used in engineering applications fall within the category of structure, characterized by microstructural details whose size
simple nonpolar materials 共Noll 1972兲, for which the stress at a ranges over many orders of magnitude. Some of these details can
given point uniquely depends on the current values and possibly be described explicitly by spatial variation of the material prop-
also the previous history of deformation and temperature at that erties. But this can never be done simultaneously over the entire
point only. Deformation is in this context characterized by the range of scales. One reason is that such a model would be pro-
deformation gradient or by an appropriate strain tensor, i.e., it is hibitively expensive for practical applications. Another, more fun-
fully determined by the first gradient of the displacement field. damental reason is that on a small enough scale, the continuum
Intuitively, it seems to be clear that the history of observable description per se is no longer adequate and needs to be replaced
by a discrete mass-point model 共or, ultimately, by interatomic
variables 共strain and temperature兲 defines the ‘‘excitation’’ of the
potentials based on quantum mechanics兲.
material point and that the corresponding ‘‘response’’ in terms of
Constructing a material model, one must select a certain reso-
stress and entropy evolution should be a unique functional of the
lution level below which the microstructural details are not ex-
local excitation at that point. However, this intuitive feeling tac-
plicitly ‘‘visible’’ to the model and need to be taken into account
itly relies on the assumption that the material can be treated as a
approximately and indirectly, by an appropriate definition of ‘‘ef-
continuum at an arbitrarily small scale. Only then the finite body
fective’’ material properties. Also, one should specify the charac-
can be decomposed into a set of idealized, infinitesimal material
teristic wave length of the imposed deformation fields that can be
volumes, each of which can be described independently as far as expected for the given type of geometry and loading. Here, the
the constitutive behavior is concerned. Of course, this does not term ‘‘wave length’’ applies not only to dynamics, where its
mean that the individual material points are completely isolated, meaning is clear, but also to statics, where it characterizes to the
minimum size of the region into which the strain can localize.
1
Walter P. Murphy Professor of Civil Engineering and Materials If the characteristic wave length of the deformation field re-
Science, Northwestern Univ., Evanston, IL 60208. E-mail: z-bazant@ mains above the resolution level of the material model, a conven-
northwestern.edu tional continuum description can be adequate. On the other hand,
2
Research Engineer, Swiss Federal Institute of Technology 共EPFL兲, if the deformation field is expected to have important components
1015 Lausanne, Switzerland. E-mail: [email protected] with wave lengths below the resolution level, the model needs to
Note. Associate Editor: Stein Sture. Discussion open until April 1, be enriched so as to capture the real processes more adequately.
2003. Separate discussions must be submitted for individual papers. To
Instead of refining the explicit resolution level, it is often more
extend the closing date by one month, a written request must be filed with
the ASCE Managing Editor. The manuscript for this paper was submitted
effective to use various forms of generalized continuum formula-
for review and possible publication on June 21, 2002; approved on June tions, dealing with materials that are nonsimple or polar, or both.
24, 2002. This paper is part of the Journal of Engineering Mechanics, Some early attempts can be traced back to the 19th century
Vol. 128, No. 11, November 1, 2002. ©ASCE, ISSN 0733-9399/2002/11- 共Voigt 1887, 1894兲, but the first effective formulation of this kind
1119–1149/$8.00⫹$.50 per page. was proposed by Cosserat and Cosserat 共1909兲. They considered

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1119

J. Eng. Mech., 2002, 128(11): 1119-1149

 material particles as objects having not only translational but also or in modeling of liquid crystals 共Oseen 1933兲, radiative transfer
rotational degrees of freedom, described by the rotation of a rigid 共Chandrasekhar 1950兲, and electric wave phenomena in the cortex
frame consisting of three mutually orthogonal unit vectors. A 共Hodgkin 1964兲. Rogula 共1965兲 proposed a nonlocal form of the
somewhat simpler concept was used by Oseen 共1933兲 and Erick- constitutive law for elastic materials.
sen 共1960兲 in their work on liquid crystals—they enriched the Nonlocal elasticity was subsequently refined by Eringen
kinematic description by the rotation of a single vector, charac- 共1966c兲, Kröner and Datta 共1966兲, Kröner 共1966, 1967兲, Kunin
terizing the orientation of the elongated axis of each crystal. After 共1966a,b, 1968兲, Edelen 共1969兲, Edelen and Laws 共1971兲, Edelen
Günther 共1958兲 had reopened the question of an oriented con- 共1971兲, Eringen et al. 共1972兲, Eringen and Edelen 共1972兲, and
tinuum and pointed out its relation to the theory of dislocations, others. These early studies, frequently motivated by homogeniza-
the old idea of the Cosserat brothers inspired a rapid development tion of the atomic theory of Bravais lattices, aimed at a better
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

leading to the couple-stress elasticity 共Mindlin and Tiersten 1962; description of phenomena taking place in crystals on a scale com-
Toupin 1962, 1964; Koiter 1964兲, theory of elasticity with micro- parable to the range of interatomic forces. They showed that non-
structure 共Mindlin 1964兲, micropolar and micromorphic theories local continuum models can approximate the dispersion of short
共Eringen and Suhubi 1964; Eringen 1964, 1966a,b兲, and multipo- elastic waves and improve the description of interactions between
lar theory 共Green and Rivlin 1964a; Green 1965兲. Nonlinear ex- crystal defects such as vacancies, interstitial atoms, and disloca-
tensions were proposed, e.g., by Lippmann 共1969兲 and Besdo tions.
共1974兲. During the last quarter of a century, it has become clear that
All these generalized Cosserat theories characterize the motion neither distributed damage in materials nor transitions to discrete
of a solid body by additional fields that are independent of the microstructural models can be adequately characterized by local
displacement field and provide supplementary information on the constitutive relations between stress and strain tensors. A great
small-scale kinematics. For example, the continuum with micro- variety of nonlocal models, involving either spatial integrals or
structure 共Mindlin 1964兲 uses for this purpose a second-order gradients of strain or internal variables, have been developed. The
tensor field that has the meaning of a ‘‘microscopic deformation present paper attempts to review the main existing models, clas-
gradient,’’ in general different from the macroscopic deformation sify them, and compare their properties. Before we focus our
gradient evaluated from the displacement field. If the microscopic attention on models of the integral type, let us discuss nonlocality
deformation gradient is restricted to orthogonal tensors, the in a more general context.
Cosserat or micropolar continuum is recovered as a special case.
On the other hand, a further generalization of the continuum with
microstructure leads to the micromorphic continuum 共Eringen Strong and Weak Nonlocality
1966b兲. In fact, the continuum with microstructure can be identi-
In solid mechanics, an integral-type nonlocal material model is a
fied with the so-called micromorphic continuum of grade 1 and
model in which the constitutive law at a point of a continuum
degree 1.
involves weighted averages of a state variable 共or of a thermody-
The micropolar continuum model can be conceived as a con-
namic force兲 over a certain neighborhood of that point. Clearly,
tinuum approximation of elastic lattices whose members possess
nonlocality is tantamount to an abandonment of the principle of
a finite bending stiffness. For instance, an orthotropic micropolar
local action of the classical continuum mechanics. A gradient-type
model was developed for large regular elastic frames and applied
nonlocal model, while adhering to this principle mathematically,
to buckling of tall buildings 共Bažant 1971; Bažant and Chris-
takes the field in the immediate vicinity of the point into account
tensen 1972a,b兲.
by enriching the local constitutive relations with the first or higher
Another important family of enriched continua retains the dis-
gradients of some state variables or thermodynamic forces. A sa-
placement field as the only independent kinematic field and im-
lient characteristic of both the integral- and gradient-type nonlo-
proves the resolution by incorporating the gradients of strain 共i.e.,
cal models is the presence of a characteristic length 共or material
higher gradients of displacement兲 into the constitutive equations.
length兲 in the constitutive relation.
Interest in such higher-grade materials or gradient theories was
The term ‘‘nonlocal’’ has in the past been used with two
stimulated by Aero and Kuvshinskii 共1960兲, Grioli 共1960兲, Raja-
senses, one narrow and one broad. In the narrow sense, it refers
gopal 共1960兲, and Truesdell and Toupin 共1960兲. These pioneers
strictly to the models with an averaging integral. In the broad
took into account only those components of the strain gradient
sense, it refers to all the constitutive models that involve a char-
that correspond to curvatures, i.e., to gradients of rotations. This
acteristic length 共material length兲, which also includes the gradi-
is equivalent to the Cosserat theory with constrained rotations, in
ent models. This broad sense stems from the realization that some
which the rotations of the rigid frame associated with each mate-
gradient models are derived as approximations to the nonlocal
rial particle are not independent but are identified with the rota-
averaging integrals, and that for all the gradient models the gra-
tion tensor resulting from the polar decomposition of the macro-
dient, in fact, includes a dependence on the immediate 共infinitely
scopic deformation gradient. Subsequently, the gradient theory
close兲 neighborhood of the point under consideration.
was extended by including the effects of the stretch gradients
A mathematical definition of nonlocality has been given, e.g.,
共Toupin 1962兲, second strain gradients 共Mindlin 1965兲, and gra-
by Rogula 共1982兲. The fundamental equations of any physical
dients of all orders 共Green and Rivlin 1964b兲. Krumhansl 共1965兲
theory can be written in the abstract form
discussed the need for higher-order displacement gradients in
continuum-based approximations of discrete lattices. Au⫽ f (1)
The last broad family of enriched continuum models is the
family to be reviewed here. It consists of nonlocal models of the where f ⫽given excitation, u⫽unknown response, and A
integral type. As early as 1893, Duhem noted that stress at a point ⫽corresponding operator 共possibly nonlinear兲 characterizing the
should, in principle, depend on the state of the whole body. Non- system. Typically, u and f are functions or distributions defined
local approaches were exploited in various branches of physical over a certain spatial domain V. Operator A is called local if it
sciences, e.g., in optimization of slider bearings 共Rayleigh 1918兲, has the following property:

1120 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 1. If two functions u and v are identical in an open set O, A Timoshenko beam can be considered as a specific one-
then their images Au and Av are also identical in O. dimensional version of a Cosserat continuum. The characteristic
Equivalently, one could say that whenever u(x)⫽ v (x) for all x in length is dictated by the square root of the ratio between the
a neighborhood of point x0 , then Au(x0 )⫽A v (x0 ). It is easily bending stiffness and the shear stiffness of the cross section. For
seen that differential operators satisfy this condition, because the a fixed shape of the cross section, the characteristic length is
derivatives of an arbitrary order do not change if the differentiated proportional to the beam depth. Note that, in this one-dimensional
function changes only outside a small neighborhood of the point description, one ‘‘point’’ of the generalized continuum corre-
at which the derivatives are taken. For example, standard one- sponds to a cross section of the beam. Therefore, only the span,
dimensional elasticity is described by the ordinary differential but not the depth, of the beam is an actual geometric dimension of
equation the model in the physical space. The beam depth is a part of the
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

generalized material model, represented by the moment-curvature

⫺ 关 E 共 x 兲 u ⬘ 共 x 兲兴 ⬘ ⫽ f 共 x 兲 (2)
relation and by the relation between the shear force and the shear
where E⫽modulus of elasticity, u⫽displacement, f ⫽body force, distortion. The presence of a characteristic length means that the
and the prime denotes differentiation with respect to the spatial solutions for different spans cannot be obtained by simple scaling
variable x. It is easily verified that the locality condition is satis- of a reference solution for a given span. Such a scaling would be
fied. Equation 共2兲 combines the strain-displacement equation, possible only if the beam depth was also scaled, but this corre-
⑀(x)⫽u ⬘ (x), equilibrium equation, ␴ ⬘ (x)⫹ f (x)⫽0, and the sponds to a change of the ‘‘material.’’
共local兲 elastic constitutive equation, ␴(x)⫽E(x)⑀(x), where ⑀ is The foregoing example clearly shows that the solution of a
the strain and ␴ is the stress. In nonlocal elasticity, the constitu- problem can be governed by the ratio of the physical dimensions
tive equation has the form of a structure to an intrinsic material length. In the present case,

冕
this material length arises from the dimensional reduction and has
⬁
␴共 x 兲⫽ E 共 x,␰ 兲 ⑀ 共 ␰ 兲 d␰ (3) its origin in the geometrical dimension that is no longer explicitly
⫺⬁ resolved by the model. In analogy to that, the material lengths that
are present in various forms of generalized continuum theories
where E(x,␰)⫽kernel of the elastic integral operator, generaliz-
arise from the homogenization procedure and have their origin in
ing the notion of the elastic modulus. The corresponding gener-
the characteristics of the heterogeneous microstructure that are no
alization of Eq. 共2兲 then reads
longer explicitly resolved.

⫺ 冋冕⫺⬁
⬁
册 ⬘
E 共 x,␰ 兲 u ⬘ 共 ␰ 兲 d␰ ⫽ f 共 x 兲 (4)
To summarize the suggested classification, continuum models
for the mechanical behavior of solids 共same as other continuum
theories兲 can be divided into
Due to the presence of a spatial integral, the locality condition is 1. strictly local models, which encompass nonpolar simple ma-
violated 关unless the elastic kernel has the degenerate form terials;
E(x,␰)⫽E(x)␦(x⫺␰), where ␦ is the Dirac distribution, in 2. weakly nonlocal models, exemplified by polar theories and
which case the local elasticity is recovered兴. gradient theories 共higher-grade materials兲; and
According to the foregoing definition, one could say that the 3. strongly nonlocal models, such as models of the integral
local theories are those described by differential equations and type.
nonlocal theories are those described by integrodifferential equa- It is worth noting that the recently emerged implicit gradient mod-
tions. But this refers to nonlocality in the narrow sense. There is els 共Peerlings et al. 1996; Geers et al. 2001; Engelen et al. 2002兲
another important aspect, related to the presence or absence of a are classified as strongly nonlocal, because they are equivalent to
characteristic length. From the mathematical point of view, the integral-type models with special weight functions used for
absence of a characteristic length is manifested by the invariance weighted averaging. In the present survey, we restrict our atten-
of the fundamental equations with respect to scaling of the spatial tion exclusively to strongly nonlocal models of the integral type.
coordinates 关for a precise definition, see Rogula 共1982兲兴. For ex-
ample, in standard linear elasticity, fundamental Eq. 共2兲 remains
valid if x is replaced by x̃⫽sx, u is replaced by ũ⫽su, and f is Integral-Type Nonlocal Models
replaced by f̃ ⫽ f /s, where s is a positive scaling parameter 共the
prime is then interpreted as the derivative with respect to x̃). This Nonlocal Elasticity
indicates that the theory does not possess any characteristic
length. A local theory invariant with respect to spatial scaling is The theories of nonlocal elasticity advanced by Eringen and
called strictly local, while a local theory not invariant with respect Edelen in the early 1970s 共Edelen et al. 1971; Eringen 1972;
to spatial scaling is called weakly nonlocal. Weakly nonlocal Eringen and Edelen 1972兲 attributed a nonlocal character to many
theories are typically described by differential equations that con- fields, e.g., to the body forces, mass, entropy, or internal energy.
tain derivatives of different orders. The coefficients multiplying They were too complicated to be calibrated and experimentally
the terms of different orders have different physical dimensions, verified, let alone to be applied to any real problems. Later sim-
and from their ratios it is possible to deduce a characteristic plifications finally led to a practical formulation in which only the
length. stress–strain relations are treated as nonlocal, while the equilib-
Typical examples of weakly nonlocal theories are the Navier– rium and kinematic equations and the corresponding boundary
Bernoulli beam on an elastic 共Winkler兲 foundation, or a Timosh- conditions retain their standard form 共Eringen and Kim 1974;
enko beam. In the former case, the characteristic length is propor- Eringen et al. 1977兲. The related variational principles have re-
tional to the flexural wave length 共of spatial oscillations produced cently been developed by Polizzotto 共2001兲. This formulation will
by a concentrated force兲, which is itself proportional to 4冑EI/c now be briefly presented, in order to introduce the basic concepts
where EI⫽bending cross-sectional stiffness and c⫽foundation and prepare a basis for extensions to the inelastic behavior, espe-
modulus 共elastic constant of the foundation兲. cially to nonlocal plasticity.

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1121

J. Eng. Mech., 2002, 128(11): 1119-1149

 A linear, small-strain, nonlocal theory of elasticity can be de- esis’’兲 and, unless there is experimental evidence to the contrary,
rived from the assumption that the elastic energy of a body V is that all stiffness coefficients decay in the same manner. This mo-
given by the quadratic functional tivates the commonly assumed form of the generalized stiffness

W⫽
1
2 冕冕
V V
⑀T 共 x兲 De 共 x,␰兲 ⑀共 ␰兲 dxd␰ (5)
De 共 x,␰兲 ⫽De ␣ 共 x,␰兲
where ␣⫽certain attenuation function. The stress–strain law then
(11)

where ⑀(x)⫽strain field and De (x,␰)⫽generalized form of the reads

elastic stiffness. The difference from the standard local theory
consists in the fact that, in general, it is impossible to express the ␴共 x兲 ⫽ 冕
V
De ␣ 共 x,␰兲 ⑀共 ␰兲 d␰⫽De 冕 V
␣ 共 x,␰兲 ⑀共 ␰兲 d␰⫽De¯⑀共 x兲
global energy as a spatial integral of an energy density that would
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

(12)
depend only on the local value of strain. Only if De (x,␰)
⫽De (x)␦(x⫺␰) Eq. 共5兲 reduces to where

W⫽
1
2 冕V
⑀T 共 x兲 De 共 x兲 ⑀共 x兲 dx⫽ 冕 V
w 关 ⑀共 x兲 ,x兴 dx (6) ¯⑀共 x兲 ⫽ 冕 V
␣ 共 x,␰兲 ⑀共 ␰兲 d␰ (13)

where w(⑀,x)⫽ 21 ⑀T De (x)⑀ is the elastic energy density. Physi- is the nonlocal strain.
cally, the generalized energy expression 共5兲 is needed if the body In an infinite isotropic body, the attenuation function depends
cannot be decomposed into infinitely small cells that interact only only on the distance between points x and ␰, and we can write
through tractions on their boundaries. This is the case, for in- ␣ 共 x,␰兲 ⫽␣ ⬁ 共 储 x⫺␰储 兲 (14)
stance, in the presence of long-range interactions among material
particles, such as atoms or molecules. To remove ambiguity, ␣ ⬁ is scaled so as to satisfy the normaliz-
Differentiating Eq. 共5兲 with respect to time, we obtain the rate ing condition
of change of elastic energy
冕 ␣ ⬁ 共 储 ␰储 兲 d␰⫽1
冕冕
(15)
1 V⬁
Ẇ⫽ ⑀˙ 共 x兲 De 共 x,␰兲 ⑀共 ␰兲 dxd␰
T
2 V V where V ⬁ ⫽entire 共one-, two-, or three-dimensional兲 Euclidean
space in which the problem is formulated. Consequently a uni-
⫹ 冕冕
1
2 V V
⑀T 共 x兲 De 共 x,␰兲 ⑀˙ 共 ␰兲 dxd␰⫽
1
2 冕冕V V
⑀T 共 x兲关 De 共 x,␰兲
form ‘‘local’’ strain field ⑀(x)⫽⑀0 is transformed into a uniform
nonlocal strain field

⫹DTe 共 ␰,x兲兴 ⑀˙ 共 ␰兲 dxd␰⫽ 冕冕 V V

⑀˙ T 共 x兲 Dsym
e 共 x,␰ 兲 ⑀共 ␰ 兲 dxd␰ (7)
¯⑀共 x兲 ⫽ 冕V⬁
␣ 共 x,␰兲 ⑀共 ␰兲 d␰⫽⑀0 冕V⬁
␣ 共 x,␰兲 d␰⫽⑀0 (16)

and the corresponding stress field is given by ␴(x)⫽De ⑀(x)

where
⫽De ⑀0 ⫽␴0 . This gives to De the physical meaning of elastic
1 stiffness under uniform straining.
e 共 x,␰ 兲 ⫽ 2 关 De 共 x,␰ 兲 ⫹De 共 ␰,x 兲兴
Dsym T
(8) The attenuation function, also called the nonlocal weight func-
tion or the nonlocal averaging function, is often assumed to have
is the symmetric part of the generalized stiffness. Note that the the form of the Gauss distribution function

冉 冊
symmetrization is carried out simultaneously with respect to the
components of the matrix De as well as to the arguments x and ␰. r2
␣ ⬁ 共 r 兲 ⫽ 共 ᐉ 冑2␲ 兲 ⫺N dim exp ⫺ (17)
We will assume that the generalized stiffness is right away de- 2ᐉ 2
fined such that it satisfies the symmetry conditions De ⫽DTe and
where ᐉ⫽parameter with the dimension of length and N dim
De (x,␰)⫽De (␰,x), and we will drop the superscript ‘‘sym.’’ The
⫽number of spatial dimensions. Function 共17兲 has an infinite
last expression in 共7兲 can be written in the form
support, which means that nonlocal interaction takes place be-
Ẇ⫽ 冕V
⑀˙ T 共 x兲 ␴共 x兲 dx (9)
tween any two points, no matter how far from each other they are.
For reasons of computational efficiency, it is more advantageous
to use attenuation functions with a finite support, e.g., the poly-
where nomial bell-shaped function 共Bažant and Ožbolt 1990兲

␴共 x兲 ⫽ 冕 V
De 共 x,␰兲 ⑀共 ␰兲 d␰ (10) ␣ ⬁ 共 r 兲 ⫽c 1⫺ 冓 r2
R2 冔 2
(18)

Eq. 共9兲 is the standard expression for the internal power delivered where the Macauley brackets 具¯典 denote the positive part, de-
by stress ␴ at strain rate ⑀˙ . Consequently, ␴ is identified as the fined as 具 x 典 ⫽max(0,x). Definition 共18兲 contains, again, a param-
stress, and Eq. 共10兲 is the constitutive equation of nonlocal elas- eter with the dimension of length R, which in this case plays the
ticity. Since the internal power expression has the standard form, role of the interaction radius, because ␣ ⬁ (r) vanishes for r⭓R.
the principle of virtual power leads to exactly the same equilib- The scaling factor c is determined from condition 共15兲 and is
rium equations and traction boundary conditions as it does in equal to 15/(16R) in one dimension, 3/(␲R 2 ) in two dimensions,
standard 共local兲 elasticity. and 105/(32␲R 3 ) in three dimensions.
For simplicity, we will restrict our attention to macroscopically There is no unique way of defining the exact form of the
homogeneous bodies. It is reasonable to assume that the interac- attenuation function in a finite body. In nonlocal elasticity, it is
tion effects decay with distance between the two points x and ␰ usually assumed that ␣(x,␰) is still given by Eq. 共14兲, regardless
共this is sometimes called the ‘‘attenuating neighborhood hypoth- of the presence of boundaries. The integral 兰 V ␣(x,␰) d␰ taken

1122 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 over the finite body V is then smaller than 1 for x close to the 共1903兲, discussed by Thomas 共1961兲, and emphasized by Sandler
boundary ⳵V, and uniform straining of the finite body does not 共1984兲, Read and Hegemier 共1984兲, and others. For a dynamic
generate a uniform stress. For materials with long-range elastic problem in one spatial dimension, they can be described as fol-
forces, this phenomenon has a clear physical explanation. For lows:
instance, consider a regular atomic lattice under constant 1. A material whose tangential stiffness becomes negative has
‘‘strain,’’ i.e., under a uniform relative increase of the interatomic an imaginary wave speed, and thus cannot propagate waves.
distances. For atoms that are sufficiently far from the boundary, 共Today we know that strain-softening concrete, in fact, can
the forces generated by the displacement of the surrounding propagate unloading waves and, due to a rate effect on crack
atoms within the interaction distance cancel out due to symmetry. growth, also loading waves of a sufficiently steep front.兲
However, if a part of the neighborhood is cut off by the boundary, 2. The dynamic initial-boundary-value problem then changes
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

some of these forces disappear and equilibrium gets disturbed. In its type from hyperbolic to elliptic and becomes ill posed,
a local continuum, it is sufficient to replace the effect of the which means that an infinitely small change in the initial
missing material by tractions on the boundary. However, in a conditions can lead to a finite change in the dynamic solu-
lattice with long-range interactions, there is a boundary layer of tion. This was analytically documented for wave propagation
thickness R that ‘‘feels’’ the absence of the material behind the in a strain-softening bar by Bažant and Belytschko 共1985兲;
boundary. To restore equilibrium, it is not sufficient to apply trac- see also Bažant and Cedolin 共1991兲, Sec. 13.1.
tion on the boundary, but additional forces are needed in the en- In multiple spatial dimensions, some wave speeds can remain real
tire boundary layer. even when the tangential stiffness tensor ceases being positive
definite, which means that stress waves can still propagate but not
in an arbitrary direction. The initial-boundary-value problem is,
Motivations of Nonlocality
again, ill posed even though, in general, it does not become ellip-
The aforementioned initial advances were motivated by devia- tic.
tions from the local constitutive models at small scales, caused Introduction of a characteristic length into the constitutive
solely by microstructural heterogeneity on the scale of the char- model 共Bažant 1976; Bažant and Cedolin 1979; Cedolin and Ba-
acteristic length. Recently, a sophisticated explanation of the need žant 1980; Pietruszczak and Mróz 1981; Bažant and Oh 1983兲,
for nonlocal terms in homogenized elastic models of random and formulation of a nonlocal strain-softening model 共Bažant
composites has been given by Drugan and Willis 共1996兲 and Lu- et al. 1984兲 and its second-gradient approximation 共Bažant
ciano and Willis 共2001兲. 1984a兲, were then shown to prevent the spurious localization of
An entirely different motivation of nonlocality—the strain- strain-softening damage 共i.e., to serve as a localization limiter兲, to
softening character of distributed damage—came to light during regularize the boundary value problem 共i.e., make it well posed兲,
the 1970s. It happened as a result of entering the computer era. and to ensure numerical convergence to physically meaningful
Finite-element programs made it suddenly feasible to simulate the solutions. In relation to nonlinear fracture mechanics, the charac-
distributed cracking observed in failure tests of concrete struc- teristic length in quasibrittle materials with distributed cracking
tures. The need to develop concrete vessels and containments for may be physically interpreted as 共or related to兲 the effective size
nuclear reactors led to lavish research funding. To approximate of the fracture process zone at the tip of a macroscopic crack.
the distributed cracking by a continuum, damage models with With respect to homogenization theory, the characteristic length
strain softening had to be introduced into finite-element codes. may be taken as equal to 共or related to兲 the size of the represen-
The first among these models was probably the smeared cracking tative volume of the material.
model of Rashid 共1968兲. The third, practically most compelling, motivation of nonlo-
The concept of strain softening violated the basic tenets of cality was the size effect 共in the present context understood as the
continuum mechanics as understood at that time, particularly the dependence of the nominal strength on the structure size兲. The
conditions of stability of material 共Drucker’s stability兲 and well existence of a nonstatistical size effect was brought to light by
posedness of the boundary value problem. Many theoreticians fracture experiments on concrete 共Walsh 1972, 1976; Bažant and
took the firm position that the concept of strain softening in any Pfeiffer 1987; Bažant and Planas 1998; Bažant 2002b兲 and dis-
form was unsound and dismissed its proponents contemptuously crete numerical simulations using, e.g., the random particle and
as diletants. The controversy, amusing in retrospect but deadly lattice models 共Bažant et al. 1990; Schlangen and van Mier 1992;
serious at that time, created passionate polemics at conferences, Schlangen 1993; Jirásek and Bažant 1995; van Mier 1997兲. In the
arguments with reviewers, and fights for money at funding agen- absence of a characteristic length, the size effect must have the
cies 共Bažant 2002a兲. A proposal for a model with strain softening form of a power law. This is the case, for example, in linear
was sure to be rejected if sent for review to these dogmatic theo- elastic fracture mechanics. The incorporation of a characteristic
reticians. But, eventually, it all had a positive effect—a compel- length is needed to describe a transitional type of size effect, in
ling motivation for nonlocal models of a new kind. which the scaling according to one power law at scales much
The finite-element simulations of failures with distributed smaller than the characteristic length transits to scaling according
共smeared兲 cracking demonstrated, and the analysis of stability and to another power law at scales much larger than the characteristic
bifurcation confirmed, that a local inelastic constitutive law with length 共Bažant 1984b; Bažant and Planas 1998; Bažant 2002b兲.
strain-softening damage inevitably leads to spurious localization Clearly pronounced size effects were also observed in tests of
of damage into a zone of zero volume 共Bažant 1976兲. This causes metals on the millimeter and micrometer scales. The results of
the numerical solution to become unobjective with respect to the experiments with bending of thin beams 共metallic films兲 共Rich-
choice of mesh and, upon the mesh refinement, to converge to a ards 1958; Stolken and Evans 1998兲, torsion of thin wires 共Mor-
solution with a vanishing energy dissipation during structural fail- rison 1939; Fleck et al. 1994兲, and microindentation 共Nix 1989;
ure. Ma and Clarke 1995; Poole et al. 1996兲 cannot be described by
Such physically absurd computational results were linked to the standard plasticity theory that lacks a characteristic length.
two problematic features, pointed out already by Hadamard Even in the elastic range, size effects on the torsional and bending

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1123

J. Eng. Mech., 2002, 128(11): 1119-1149

Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. Idealized crack array and energy release zone of one crack,
Fig. 1. Left: representative volume of material used for nonlocal analyzed in Bažant 共1991兲
strain averaging; the scatter of microstresses and smoothed macros-
tress profile demonstrate how the stress value at the center of repre- tion of that volume 共Bažant 1987, 1991兲. The size of this
sentative volume differs from the stress corresponding to average volume is determined by the size of the microcracks and
strain over this volume. Top right: averaging zone when the theoret- their average spacing, which may, but need not, be related to
ical nonlocal neighborhood of a point protrudes through the boundary the size of the inhomogeneities in the material.
of solid. Bottom right: bell-shaped weight function for nonlocal av- 5. Microcrack interaction, particularly the fact that one micro-
eraging integral and its relation to characteristic length ᐉ. crack may either amplify the stress intensity factor of an-
other adjacent microcrack or shield that crack, depending on
the orientations of the microcracks, the orientation of the
stiffness contradicting the standard elasticity theory were mea- vector joining their centers, and the microcrack sizes. This
sured for porous materials such as foams and human bones leads to a different kind of nonlocality that is not described
共Lakes 1986兲. explicitly by a spatial averaging integral but implicitly by an
To sum up, the development of nonlocal models was histori- integral equation with a kernel of zero average 共Bažant 1994;
cally motivated by Bažant and Jirásek 1994; Jirásek and Bažant 1994兲.
1. the need to capture small-scale deviations from local con- 6. In the case of metal plasticity, the density of geometrically
tinuum models caused by material heterogeneity; necessary dislocations in metals, whose effect, after con-
2. the need to achieve objective and properly convergent nu- tinuum smoothing, naturally leads to a first-gradient model
merical solutions for localized damage; 共Fleck and Hutchinson 1993; Gao et al. 1999; Huang et al.
3. the need to regularize the boundary value problem 共prevent 2000; Fleck and Hutchinson 2001兲.
ill posedness兲; and 7. Weibull-type extreme value statistics of quasibrittle failure.
4. the need to capture size effects observed in experiments and As recently realized, without assuming the failure probabil-
in discrete simulations. ity at a point of the material to depend on the average strain
from a finite neighborhood of the point rather on the con-
Causes of Nonlocality tinuum stress at that point, a Weibull-type weakest link
theory of quasibrittle structural failure runs into some para-
The physical causes of nonlocality, some of which have already doxical situations or incorrect predictions 共Bažant and Xi
been mentioned among the historical motivations, may be sum- 1991; Bažant and Novák 2000a,b,c兲, which cannot be
marized as follows: avoided without resorting to a nonlocal probabilistic model
1. Heterogeneity of microstructure and its homogenization on a 共see the section on nonlocal probabilistic models of failure兲.
small scale on which the smoothed strain field cannot be
considered as uniform. What matters for the macroscopic
stress 共averaged over the representative volume兲 is not the Nonlocal Averaging Operator
strain value at the center point but the average strain value Generally speaking, the nonlocal integral approach consists in
within the representative volume, which can be very differ- replacing a certain variable by its nonlocal counterpart obtained
ent 共Fig. 1, left兲. The heterogeneity may, for instance, be by weighted averaging over a spatial neighborhood of each point
caused by, and the characteristic length governed by, the under consideration. If f (x) is some ‘‘local’’ field in a solid body
grain 共inclusion兲 size, the pore size, or the size of crystals in occupying a domain V, the corresponding nonlocal field, labeled
a metal. In concrete, what is decisive is the maximum aggre- by an overbar, is defined by

冕
gate size and spacing.
2. Homogenization of regular or statistically regular lattices or f̄ 共 x兲 ⫽ ␣ 共 x,␰兲 f 共 ␰兲 d␰ (19)
frames, for example, those to be used for large planned space V
structures or for very tall buildings 共this may, of course, be
where ␣(x,␰)⫽chosen nonlocal weight function. In applications
regarded as a special case of heterogeneity兲.
to softening materials, it is often required that the nonlocal opera-
3. The fact that distributed cracking is physically observed yet
impossible to simulate numerically with local continuum tor should not alter a uniform field, which means that the weight
models, as already elaborated on. function must satisfy the normalizing condition
4. The fact that the growth of a microcrack is not decided by
the local stress or strain tensor at the continuum point corre-
sponding to the microcrack center but by the overall energy
冕 V
␣ 共 x,␰兲 d␰⫽1 ᭙x苸V (20)

release from a finite volume surrounding the whole microc- In an infinite, isotropic, and homogeneous medium, the weight
rack 共Fig. 2兲. The growth depends on the average deforma- function depends only on the distance r⫽ 储 x⫺␰储 between the

1124 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 ‘‘source’’ point, ␰, and the ‘‘receiver’’ point, x. So, we may write
␣(x,␰)⫽␣ ⬁ ( 储 x⫺␰储 ) where ␣ ⬁ (r) is typically chosen as a non-
negative bell-shaped function 共17兲 or 共18兲, monotonically de-
creasing for r⭓0 共Fig. 1, bottom right兲. The smallest distance
between points x and ␰ at which the interaction weight ␣ ⬁ ( 储 x
⫺␰储 ) vanishes 共for weight functions with a bounded support兲 or
becomes negligible 共for weight functions with an unbounded sup-
port兲 is called the nonlocal interaction radius R. The interval,
circle, or sphere of radius R, centered at x, is called the domain of
influence of point x.
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

In the vicinity of the boundary of a finite body, it is simply

assumed 共without any deep theoretical support兲 that the averaging
is performed only on the part of the domain of influence that lies
within the solid 共Fig. 1, top right兲. To satisfy condition 共20兲, the Fig. 3. Imbrication 共overlapping兲 of finite elements, approximately
weight function is usually defined as equivalent to averaging of total strain with uniform weight function
共Bažant et al. 1984兲
␣ ⬁ 共 储 x⫺␰储 兲

冕
␣ 共 x,␰兲 ⫽ (21)
V ␣ ⬁ 共 储 x⫺␨储 兲 d␨
However, this ‘‘imbricate’’ concept caused unwanted prob-
However, this modification breaks the symmetry of the weight lems. A nonlocal continuum in which the total strain is nonlocal
function with respect to the arguments x and ␰. In certain types of can exhibit zero energy modes of instability 共it appears that simi-
nonlocal theories it is desirable to work with a symmetric weight lar problems could have afflicted also various previous nonlocal
function. Polizzotto 共2002兲 and Borino et al. 共2002兲 proposed an- elastic models and went undetected, apparently for lack of finite-
other modified weight function, element implementation兲. That such instabilities exist becomes

冋 冕 册
clear upon noting that the one-dimensional integral equation
␣ 共 x,␰兲 ⫽␣ ⬁ 共 储 x⫺␰储 兲 ⫹ 1⫺ ␣ ⬁ 共 储 x⫺␨储 兲 d␨ ␦ 共 x⫺␰兲 (22) ¯⑀ (x)⬅ 兰 x⫺R
x⫹R
⑀(s) ds⫽0 is satisfied by any function
V
␲n 共 s⫹c 兲
which preserves symmetry and satisfies condition 共20兲. ⑀ 共 s 兲 ⫽A sin (23)
R
In computer programs, the nonlocal average at x is calculated
as a weighted sum over the values at all the finite-element inte- where n⫽arbitrary positive integer, and A and c⫽arbitrary real
gration points ␰ lying within the nonlocal interaction radius R. constants. In general, it was shown that such instabilities are
One inevitable penalty of nonlocal averaging is that the band- avoided if and only if the Fourier transform of the weight function
width of the stiffness matrix gets increased. This increases the ␣ ⬁ (r) is everywhere positive 共Bažant and Chang 1984兲. This is
relative attractiveness of explicit finite-element schemes, which not the case for a uniform weight function 共and thus not for the
do not necessitate the assembly and decomposition of the stiffness ‘‘imbricate’’ continuum model兲. Neither is this the case for a tri-
matrix. For nonlocal damage models, it is nevertheless possible to angular weight function or a truncated polynomial function 共18兲.
construct the consistent tangent stiffness matrix 共Huerta and The Gaussian function 共17兲 has a positive Fourier transform.
Pijaudier-Cabot 1994兲 and use it efficiently in implicit finite- But, since the tail of its Fourier transform approaches zero, the
element computations 共Jirásek and Patzák 2002兲. situation is close to instability. A robust remedy is to add a mul-
tiple of the Dirac distribution to the weight function, which is
equivalent to an ‘‘overlay’’ 共parallel coupling兲 with an elastic
Original Nonlocal Model for Strain Softening and Its local continuum. But, then, full strain softening down to a zero
Limitations stress becomes impossible.
In the 1980s, nonlocal models were extended to inelastic materi- So, this original nonlocal model for regularizing strain-
als. Eringen developed his nonlocal formulation of isotropically softening problems was found usable only for a material that
hardening plasticity in strain space 共Eringen 1981兲, perfect plas- suffers merely partial softening damage followed by rehardening,
ticity with associated flow, and deformation theory of plasticity but not in general situations. Therefore, other remedies were
共Eringen 1983兲. Subsequently, Eringen and Ari 共1983兲 applied sought, as described next.
these models to simulation of the yielding zone at the fracture
front.
The nonlocal averaging concept was also applied to a strain- Nonlocal Damage and Smeared Cracking
softening damage model by Bažant et al. 共1984兲 in order to regu-
larize the boundary value problem and prevent spurious depen-
Nonlocal Damage Models
dence of the process zone width and of the energy dissipation on
mesh refinement. Initially, the averaging operator was applied to Understanding the source of the aforementioned problems sug-
the total strain tensor ⑀. With a uniform weight function, this gested the remedy—the instability modes obviously cannot arise
model could be easily implemented in a finite-element code by if the nonlocal averaging is applied to variables that can never
imbricating 共i.e., overlapping兲 the finite elements in the manner of decrease. Such variables, for example, include the damage vari-
roof tiles 共‘‘imbrex’’ in Latin兲, as shown in Fig. 3. With the con- able ⍀ or the maximum level of damage energy release rate Y max
dition that the element size be kept constant regardless of mesh in continuum damage mechanics, or the cumulative plastic strain
refinement, the elements themselves performed the strain averag- ␬ in strain-softening plasticity.
ing. The characteristic length ᐉ was then proportional to the fixed This idea proved successful. It was first applied to continuum
element size. damage mechanics, exemplified by the simple isotropic damage

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1125

J. Eng. Mech., 2002, 128(11): 1119-1149

 model with one scalar damage variable 共e.g., Lemaitre and
Chaboche 1990兲. In its local form, this model can be described by
the following set of equations:
␴⫽ 共 1⫺⍀ 兲 De ⑀ (24)
⍀⫽␻ 共 Y max兲 (25)
Y max共 t 兲 ⫽max Y 共 ␶ 兲 (26)
␶⭐t

Y ⫽ 21 ⑀T De ⑀ (27)
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

In the above, ␴⫽column matrix of six stress components, ⑀

⫽column matrix of six engineering strain components, De
⫽elastic material stiffness matrix, and ⍀⫽damage variable that
grows from zero 共virgin state兲 to one 共fully damaged state兲 de-
pending on Y max , which is the maximum value of the damage
energy release rate Y ever attained in the previous history of the
material up to the current state. Of course, Y is not a rate in the
sense of a derivative with respect to time. It equals minus the
derivative of the free-energy density ␳␺(⑀,⍀)⫽(1⫺⍀)⑀T De ⑀/2
with respect to the damage variable ⍀, and so it represents the
‘‘rate’’ at which energy would be released during 共artificially in-
duced兲 damage growth at constant strain and temperature.
The monotonically increasing function ␻ in Eq. 共25兲 controls
the evolution of damage and thus affects the shape of the stress–
strain curve. It is usually designed such that ⍀⫽0 as long as Y max
remains below a certain threshold value, Y 0 . In view of the afore-
mentioned conclusion about the proper nonlocal approach, this
local damage evolution law was adapted to a nonlocal form in Fig. 4. Profiles of incremental strain per unit stress increment ob-
either of the two following ways 共Pijaudier-Cabot and Bažant tained in nonlocal numerical simulation of a tensioned bar: 共a兲 com-
1987; Bažant and Pijaudier-Cabot 1988兲, which represent averag- parison of two types of averaging; 共b兲 convergence as the number of
ing of the damage energy release rate or of the damage variable: elements is increased 共after Bažant and Pijaudier-Cabot 1988兲.

⍀⫽␻ 共 Ȳ max兲 , or ⍀⫽␻ 共 Y max兲 (28)

The convergence and stability of this formulation was verified in 5. The 共local兲 damage evolution law ⍀⫽␻(Y max) is constructed
various ways; e.g., by refining the mesh for a uniaxially stretched such that the response remains linear up to the peak stress and the
softening bar. The numerical solutions for a progressively increas- softening curve is exponential, asymptotically approaching the
ing number of finite elements are plotted in Fig. 4. As seen, the horizontal axis. The dashed curve in Fig. 5 corresponds to the
profiles of strain increment throughout the softening zone con- unstable solution for which the strain remains uniform. The actual
verge as the number of elements along the bar is increased from stable solutions are characterized by a nonuniform strain distribu-
12 to 100. Also, the averaging of damage and of energy release tion, with strain increments localized into a finite interval, the size
rate give, in this case, very similar results. of which is controlled by the characteristic length 共see Fig. 4兲.
The latter, though, is not true for large postpeak deformations, The initial response, right after the onset of localization, is about
as demonstrated by Jirásek 共1998b兲. At very large extensions of the same for all the models considered here. At later stages of
the bar, a complete fracture must be simulated, which means that softening, only the formulations labeled ␻(⑀ ¯ ), Ȳ , and ¯␥ give a
the stress must be reduced to zero. Jirásek 共1998b兲 showed that reasonable behavior, with the residual strength approaching zero
this is not true for some types of nonlocal averaging, and found as the applied elongation is increased 关Fig. 5共a兲兴. The other for-
that averaging of different variables gives rather different re- mulations lead to locking effects and sometimes fail to converge
sponses. The nonlocal damage formulations that he considered 关Fig. 5共b兲兴. Thus, it can be concluded that the complete fracture is
are summarized in Table 1. The central column presents in a correctly reproduced by models that average the equivalent strain,
compact form the nonlocal stress–strain law for the isotropic
damage model whose local version is described by Eqs. 共24兲–
共27兲. The right column shows a possible generalization to aniso- Table 1. Overview of Nonlocal Damage Formulations
tropic damage 共top part兲 or to a completely general inelastic Formation Isotropic damage model General model
model 共bottom part兲. Beside the symbols already defined, the fol-
lowing notations are used: ␥⫽⍀/(1⫺⍀)⫽compliance variable, ␻共␧¯兲 ␴⫽ 关 1⫺␻(Y ¯␧ ))]De ␧ ␴⫽Ds (␧
¯ )␧
Ds ⫽damaged 共secant兲 stiffness matrix, Du ⫽unloading stiffness Ȳ ␴⫽ 关 1⫺␻(Y ␧))]De ␧ ␴⫽Ds (⍀(Y (␧)))␧
matrix, ⍀⫽damage tensor, Y⫽tensor of damage energy release ¯␧ ␴⫽ 关 1⫺␻(Y ␧))]De ␧ ␴⫽Ds (␧)␧
rates work-conjugate to ⍀, Ce ⫽D⫺1 e ⫽elastic compliance matrix, ¯␥ ␴⫽ 关 1⫹␥(Y ␧))] ⫺1 De ␧ ␴⫽ 关 Ce ⫹Ce (␧) 兴 ⫺1 ␧
Ci ⫽inelastic compliance matrix, and s⫽De ⑀⫺␴⫽inelastic s̄ ␴⫽De ␧⫺␻(␧)De ␧ ␴⫽De ␧⫺s(␧)
stress. An overdot denotes differentiation with respect to time. ˙ ⫽(1⫺␻)De ␧˙ ⫺␻
␴ ˙ De ␧
¯
⌬s ␴⫽Du ␧˙ ⫺ṡ(␧,␧˙ )
The load-displacement diagrams generated in a uniaxial tensile
s共␧¯兲 ␴⫽De ␧⫺␻(␧
¯ )De¯␧ ␴⫽De ␧⫺s(␧
¯)
test by different nonlocal damage formulations are shown in Fig.

1126 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

Fig. 5. Comparison of postpeak load-deflection curves of a tensioned bar calculated for nonlocal damage with averaging applied to different
variables: 共a兲 nonlocking formulations ␻(⑀
¯ ), Ȳ and ¯␥ ; and 共b兲 locking formulations ␻
¯ , s̄, ⌬s, and s(⑀
¯ ).

the energy release rate, or the compliance variable. The evalua- two internal lengths leading to two nonlocal averages of the strain
tion of the inelastic stress from the nonlocal strain, the same as tensor, driving the damage evolution in tension and in compres-
the averaging of the damage variable, inelastic stress, or inelastic sion. Jirásek 共1999兲 started from the microplane damage frame-
stress increment, leads to spurious residual stresses and to an work established by Carol and Bažant 共1997兲, which exploits the
expansion of the softening zone across the entire bar. Models principle of energy equivalence 共Cordebois and Sidoroff 1979兲.
based on averaging of inelastic stress will be addressed in more He developed an anisotropic damage model for tensile failure of
detail in the section on nonlocal adaptation of general constitutive concrete and regularized it by averaging the compliance param-
models. eters related to microplanes of different orientations, which cor-
The basic model with damage evolution driven by the damage responds to an anisotropic extension of formulation ¯␥ . Fish and
energy release rate 共27兲 is simple and appealing from the theoret- Yu 共2001兲 derived a nonlocal damage model for composites by a
ical point of view, since it can be formulated within the frame- multiscale asymptotic analysis of the damage phenomena occur-
work of generalized standard materials 共Halphen and Nguyen ring at the micro-, meso-, and macrolevel. Huerta et al. 共1998兲
1975兲. However, it is not suitable for quasibrittle materials, be- developed a mesh-adaptive technique for nonlocal damage mod-
cause it gives the same response in tension and in compression. els, and Rodrı́guez-Ferran and Huerta 共2000兲 performed adaptive
To emphasize the effect of tension on the propagation of cracks, simulations of concrete failure using a residual-type error estima-
Mazars 共1984兲 proposed to link the damage to the so-called tor 共Dı́ez et al. 1998兲.
equivalent strain, defined as the norm of the positive part of the
strain tensor. He developed an isotropic damage model for con-
crete with two damage parameters, ␻ t and ␻ c , which correspond Nonlocal Smeared Cracking
to uniaxial tension and uniaxial compression, respectively, and The nonlocal concept has further been applied to the smeared
both depend on the maximum previously reached value of the cracking models widely used for concrete structures and rocks.
equivalent strain. For a general stress state, the actual damage Such models transform the cumulative effect of microcrack
parameter is interpolated according to the current values of prin- growth and coalescence into additional inelastic strain due to
cipal stresses. The nonlocal formulation of Mazars’ model was cracking. Adding the smeared cracking strain to the elastic strain,
refined by Saouridis 共1988兲 and Saouridis and Mazars 共1992兲, one obtains the total strain. A secant or tangential compliance
following the basic idea of Pijaudier-Cabot and Bažant 共1987兲. tensor of the cracked material, which is orthotropic, can be easily
The averaged quantity was the equivalent strain, which corre- obtained. If the damage remains distributed, the cracking strain
sponds to a natural generalization of formulation Ȳ . corresponds to the average of crack opening divided by the spac-
A number of nonlocal damage formulations appeared in the ing between parallel cracks. But, the model can be used even after
literature during the last decade. For instance, di Prisco and Maz- localization of damage into large macroscopic cracks. To obtain
ars 共1996兲 improved the nonlocal version of Mazars’ model by an objective description, it is necessary either to relate the soft-
introducing irreversible strains and an additional internal variable ening part of the stress–strain law to the numerically resolved
controlling the volumetric expansion of concrete in compression. width of the fracture process zone 共which depends on the finite-
The model was further refined by Ferrara 共1998兲, who considered element size and tends to zero as the mesh is refined兲, or to
the characteristic length as a variable parameter that depends on reformulate the model as nonlocal and enforce a mesh-
the current level of damage. Valanis started from his anisotropic independent width of the process zone by introducing a charac-
damage theory based on the integrity tensor 共Valanis 1990兲 and teristic length.
reformulated it as nonlocal, with the damage rate dependent on The smeared cracking model has two variants: 共1兲 the fixed
the current local damage, strain, and positive part of the nonlocal crack model, in which the crack orientation is fixed when the
strain rate 共Valanis 1991兲. Kennedy and Nahan 共1996兲 proposed a maximum principal stress first attains the strength limit 共Rashid
nonlocal anisotropic damage model 共with fixed axes of material 1968; de Borst 1986兲; and 共2兲 the rotating crack model 共Cope
orthotropy兲 for composites and applied it to the failure analysis of et al. 1980; Gupta and Akbar 1984; Rots 1988兲, in which the
laminated shells 共Kennedy and Nahan 1997兲. Their approach is an crack orientation is rotated so as to always remain perpendicular
anisotropic extension of the formulation ␻(⑀ ¯ ). A different exten- to the maximum principal strain direction. The latter concept,
sion of this formulation was used by Comi 共2001兲, who defined which usually gives more realistic results, does not mean that the

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1127

J. Eng. Mech., 2002, 128(11): 1119-1149

 In terms of the nonlocal damage formulations summarized in
Table 1, the Bažant–Lin approach corresponds to formulation ¯␥
while the Jirásek–Zimmermann approach corresponds to formu-
lation ␻(⑀¯ ). When applied directly to the standard rotating crack
model, the latter formulation leads to instabilities due to negative
shear stiffness coefficients. A stable behavior is guaranteed for a
modified rotating crack model with transition to scalar damage
共Jirásek and Zimmermann 1998b兲.
The following example, taken from Jirásek and Zimmermann
共1997兲, demonstrates the ability of the nonlocal rotating crack
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

model to reproduce a relatively complex curved fracture pattern.

In a series of experiments, Nooru-Mohamed 共1992兲 tested the
Fig. 6. Size effect obtained with the nonlocal smeared cracking double-edge-notched 共DEN兲 specimen shown in Fig. 7共a兲, which
model, for two types of strain-softening law 共after Bažant and Lin can be subjected to a combination of shear and tension 共or com-
1988a兲 pression兲. Nooru-Mohamed performed the experiments for a
number of loading paths, some of them even nonproportional.
One of the most interesting loading scenarios is path 4c, which
cracks would actually rotate. Rather, it means that cracks of many
produces curved macroscopic cracks 关Fig. 7共b兲兴. The specimen is
orientations exist; cracks of some orientation close and cracks of
first loaded by an increasing ‘‘shear’’ force, P s , while keeping the
another orientation open, with the effect that the orientation of the
‘‘normal’’ force, P, at zero. After reaching the peak of the P s – ␦ s
dominant cracks rotates. In the case of the fixed crack model, one
curve, the type of loading changes. From that moment on, P s is
needs to cope with the problem of a secondary crack system
kept constant and the normal displacement ␦ is increased, which
crossing at some angle the system formed previously 共de Borst
results into a nonzero reaction P.
1986兲.
The aforementioned experiment was simulated using the non-
Since smeared cracking models take into account the orienta-
local rotating crack model with transition to scalar damage. The
tion of cracks, they automatically reflect the crack-induced aniso-
meshes were constructed following a pseudoadaptive technique
tropy of the material. It is possible to interpret them formally as
proposed by Jirásek and Zimmermann 共1997兲, starting from a
special anisotropic damage models.
mesh of uniform density. A sequence of three progressively re-
A nonlocal generalization of the rotating crack model, in
fined meshes along with the corresponding process zones is
which the cracking strain is averaged using the nonlocal operator
shown in Fig. 8, in which the light regions indicate high levels of
defined in Eq. 共19兲, was investigated by Bažant and Lin 共1988a兲.
damage. The simulated process zone matches the experimentally
It was shown that the size effect obtained by finite-element simu-
observed cracks very well.
lations approximately follows Bažant’s size effect law 共Bažant
1984b兲. Fig. 6 illustrates the differences in size effect for two
different types of the strain-softening law 共linear and exponen- Alleviation of Mesh Orientation Bias for Crack
tial兲. Propagation Direction
An alternative nonlocal formulation of the rotating crack
model can be based on the evaluation of the damaged material In connection with smeared cracking, it is appropriate to point out
stiffness matrix from the nonlocal average of the total strain one advantageous side benefit brought about by the nonlocal av-
共Jirásek and Zimmermann 1998b兲. The stress is then computed as eraging concept—the alleviation of mesh orientation bias. Fig. 9
the product of the nonlocally evaluated stiffness with the local shows a beam analyzed with an aligned mesh and with a deliber-
strain 共this ensures that the response in the elastic range is local兲. ately rotated mesh. The local cracking models or cohesive models

Fig. 7. Double-edge-notched specimen and observed cracks 共after Nooru-Mohamed 1992兲

1128 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 Fig. 10. Simulated crack pattern in the central part of a notched
three-point bend specimen: 共a兲 standard rotating crack model 共local兲,
共b兲 rotating crack model with transition to scalar damage 共local兲, and
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

共c兲 nonlocal rotating crack model with transition to scalar damage

共after Jirásek and Zimmermann 1998b兲.

specimen simulated with smeared crack models on a skewed

mesh. Due to symmetry, the actual crack trajectory should be
straight and vertical. Of course, the real crack path is tortuous
and, in one single experiment, may deviate from this ideal trajec-
tory. However, the computational simulation is supposed to repro-
duce the mean trajectory, averaged over a large number of experi-
ments, which is no doubt expected to lie on the axis of symmetry.
The local version of the standard rotating crack model exhibits
strong directional bias—the crack band propagates along the
mesh lines over the entire depth of the specimen 关Fig. 10共a兲兴. It is
Fig. 8. Simulation of curved crack propagation in a double-edge-
interesting to note that a partial improvement is achieved already
notched specimen using a pseudoadaptive technique: Sequence of by means of the local version of the modified rotating crack
meshes and fracture process zones 共after Jirásek and Zimmermann model with transition to scalar damage 关Fig. 10共b兲兴. This can be
1997兲. explained by the fact that, for the standard rotating crack model,
the crack trajectory shown in Fig. 10共b兲 would lead to stress lock-
ing 共Jirásek and Zimmermann 1998a兲, which means that it is
on a fixed mesh would give an inclined direction of propagation easier for the macroscopic crack to keep propagating along the
for the rotated mesh. Simply, the cracks prefer to run in the di- mesh lines.
rection of mesh lines. The nonlocal model, on the other hand, When the modified rotating crack model is reformulated as
gives a nearly vertical propagation of the damage band even for nonlocal, the results improve further, even for a rather coarse
the rotated mesh, provided that the elements are no larger than mesh for which the nonlocal interaction radius used in the present
about 31 of the process zone width 共Bažant and Lin 1988a兲. The example is only 1.25 times the element size. The resulting mac-
1
3 rule for avoiding the mesh orientation bias seems to be gener- roscopic crack trajectory only slightly deviates from the axis of
ally applicable to all the models with nonlocal averaging. symmetry, despite the skewed character of the mesh 关Fig. 10共c兲兴.
The nonlocal averaging concept seems to be the most effective The directions of ‘‘local cracks’’ 共marked by dark rectangles at
way to avoid mesh-induced directional bias, provided, of course, individual Gauss points兲 oscillate and are not aligned with the
that the use of small enough finite elements is feasible. Fig. 10 overall crack trajectory. However, this is quite natural because
shows the fracture process zone in a notched three-point bend these local cracks, defined just for the purpose of visualization of

Fig. 9. Study of mesh-orientation bias with nonlocal smeared cracking model of Bažant and Lin 共1988a兲

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1129

J. Eng. Mech., 2002, 128(11): 1119-1149

Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

Fig. 11. Microcrack interaction approach: stress on cracks ␮ and ␭ caused by unit pressure on the faces of crack ␯, and superposition and
iteration approach to a system of interacting cracks

the results, are computed from the local inelastic strains. At very an infinite space containing only one pressurized open crack. This
late stages of the stiffness degradation process, the local strains solution is well known, not only for two but also for three dimen-
localize in a single layer of elements, even though the model is sions 共Fabrikant 1990兲.
nonlocal. This cannot be judged as a deficiency of the model. Based on this kind of solution, it was shown 共Bažant 1994; see
What matters is that the load-displacement diagram and the over- also Bažant and Planas 1998, Sec. 13.3兲 that the nonlocal incre-
all direction of the cracking band are reproduced correctly, re- mental constitutive law should have the general form
gardless of the mesh size and orientation. Still better results can
⌬␴⫽De ⌬⑀⫺⌬s̃ (29)
be expected for a refined mesh with elements several times
smaller than the interaction radius. in which the nonlocal inelastic stress increment ⌬s̃ is the solution
of the Fredholm integral equation
Oriented Long-Range Nonlocality Due to Microcrack
Interactions ⌬s̃共 x兲 ⫽⌬s̄共 x兲 ⫹ 冕
V
⌳ 共 x,␰兲 ⌬s̃共 ␰兲 d␰ (30)
The models discussed so far introduce ad hoc nonlocal averaging,
without any physical justification. Bažant 共1994兲 proposed a averaging ⫹ crack interactions
physical motivation of nonlocal damage, based on a microme- The inelastic stress is understood as the stress drop due to
chanical analysis of microcrack interactions. He exploited the so- the presence of cracks, as compared to the elastic stress that
lution procedure for a statistically homogeneous system of ran- would be induced in an undamaged material by the same strain;
dom cracks in an elastic matrix, developed in detail by Kachanov see Fig. 12.
共Kachanov 1985; Kachanov 1987兲. The solution is a sum of a Equation 共30兲 is justified by the fact that its approximation by
trivial solution of an elastic solid with all the cracks imagined a discrete sum over ⌬s̄ values at the random crack centers reduces
closed 共as if ‘‘glued’’兲, and the solution for the case in which the this integral equation to a matrix equation for the solution of the
cracks are ‘‘unglued’’ and their faces are loaded by releasing the many-crack system in Fig. 11. The first term on the right, which
stresses transmitted across the glued cracks 共Fig. 11兲. Only the describes the standard nonlocal averaging according to Eq. 共19兲,
second case matters for the stress intensity factors of the cracks. is justified by material heterogeneity and by the finiteness of the
The solution for many cracks can be constructed by an iterative energy release zone of a crack. The second term, which describes
relaxation scheme on the basis of the known solution of the stress microcrack interactions, contains the microcrack interaction func-
transmitted across the plane of an arbitrary closed 共glued兲 crack in tion ⌳(x,␰), which represents statistical continuum smearing of

Fig. 13. Parameters describing relative position of two interacting

Fig. 12. Stress decomposition into elastic and inelastic parts cracks

1130 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 discrete coefficients ⌳ ␮␯ of the aforementioned matrix equation, Measurement of Material Characteristic Length for
each of which is defined as the stress across a glued crack ␮ Nonlocal Damage
produced by a unit pressure applied on the faces of crack ␯ 共Fig.
11兲. The domain of the integral must be taken as the entire body The characteristic length ᐉ is a material parameter controlling the
spread of the nonlocal weight function. It cannot be directly mea-
volume V, because the decay of function ⌳ with distance is rather
sured but can be indirectly inferred by inverse analysis of test
slow.
results. However, a reliable identification procedure can be based
The microcrack interaction function ⌳ depends not only on the
only on experiments that are sufficiently sensitive to the nonlocal
distance r⫽ 储 x⫺␰储 between points x and ␰, but also on the ori-
aspects of the material behavior.
entation of dominant cracks at those points. The dominant crack
At this point, it is important to note that there is a subtle
is assumed to be perpendicular to the maximum principal stress
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

difference between the notions of an intrinsic material length and

direction. In two dimensions, the interaction factor a nonlocal characteristic length. The latter is a special case of the
k共 r 兲 former, because nonlocal averaging is only one of the possible
⌳ 共 r,␪,␺ 兲 ⬇⫺ 关 cos 2␪⫹cos 2␺⫹cos 2 共 ␪⫹␺ 兲兴 (31) enrichments that introduce a length parameter into the continuum
2ᐉ 2
theory. Certain other enrichments, e.g., by higher-order gradients
is a function of the distance r, of the angle ␪ between the radial or independent microrotation fields, can be classified as weakly
coordinate axis r and the maximum principal stress direction at nonlocal theories and have similar regularizing effects on the so-
the source point ␰, and of the angle ␺ between the radial coordi- lution. When properly formulated, all the nonlocal theories in the
nate axis r and the maximum principal stress direction at the broad sense lead to continuous strain profiles and to process zones
receiver point x 共Fig. 13兲. Parameter ᐉ is a length parameter re- 共regions of localized strain兲 of nonzero width and length.
flecting the dominant spacing of interacting microcracks. A simi- On the other hand, there exist special techniques such as the
crack band model 共Bažant 1982; Bažant and Oh 1983兲, to be
lar but more complicated approximate expression was proposed
discussed in detail later on, for which the global solution charac-
for function ⌳ in three dimensions 共Bažant 1994兲.
teristics 共such as the load-displacement diagram and the energy
The far-away asymptotic form of radial function k(r) for r
dissipated during failure兲 remain objective with respect to the
Ⰷᐉ was derived from the remote asymptotic field of a single
mesh refinement but the process zone eventually collapses into a
pressurized crack in an infinite space. It turns out that, asymptoti-
surface or curve of vanishing thickness. Since the process zone
cally for r→⬁, k(r) is proportional to r ⫺d where d is the number
keeps a finite length, such models still carry information about the
of spatial dimensions 共2 or 3兲. As the simplest match between this
intrinsic material length and are capable of describing size effects
asymptotic field and the value 0 required for r→0, Bažant 共1994兲
of the transitional type. This material length is, however, not re-
proposed to use in two dimensions the expression
ally a nonlocal characteristic length. It is present, for instance, in

k共 r 兲⫽ 冉 ␬ᐉr
r 2 ⫹ᐉ 2 冊 2
(32)
models that work simultaneously with a fracture energy and a
material strength 共such as the crack band model兲. A classical ex-
ample is provided by Irwin’s parameter 共Irwin 1958兲,
An interesting point to note is that the asymptotic decay of k(r)
K 2c
with r is very slow, in fact, so slow that the integral in Eq. 共30兲 is r p⫽ (34)
at the limit of integrability for the case of an infinite solid. An- ␲␴ 20
other point to note is that the integral in Eq. 共30兲 does not repre- which was introduced to characterize the length of the plastic
sent nonlocal averaging because the weight function ⌳ does not zone around a crack tip in elastoplastic fracture mechanics 关for
satisfy the normalizing condition 共20兲. In fact, for a body sym- concrete, it was adopted by Hillerborg et al. 共1976兲兴. Here, K c is
metric with respect to the origin we would get the fracture toughness 共critical value of the stress intensity factor兲
and ␴ 0 is the yield stress.
冕 V
⌳ 共 0,␰兲 d␰⫽0 (33) The foregoing analysis suggests that the nonlocal models have,
in fact, two parameters with the dimension of length, one of them
characterizing the length and the other the width of the process
For angles 兩 ␪ 兩 ⬎56° 共in two dimensions兲, the normal stress zone. They are, in general, independent. Any identification proce-
component in the direction of the normal to a crack at ␰ is tensile dure that extracts from the given experimental data only a single
and tends to amplify the growth of a crack at x. For 兩 ␪ 兩 ⭐56°, it length parameter and relates it to the nonlocal characteristic
tends to shield the other crack from stress and thus inhibit its length is in reality based on the tacit assumption that the two
growth. Thus, a physically important feature of the crack interac- length parameters are somehow linked, e.g., that their ratio is
tion function is that it can distinguish between remote amplifica- fixed. For a limited class of materials, such an assumption can be
tion and shielding sectors of a crack, which is ignored in the reasonable, but it can hardly be considered as a general rule. A
standard approach. reliable and fully general identification procedure should be ca-
Eq. 共31兲 was introduced by Ožbolt and Bažant 共1996兲 into a pable of extracting two independent length parameters, but for
nonlocal finite-element code with the microplane constitutive this it is necessary to have a sufficiently rich set of experimental
model for concrete. Eq. 共30兲 was solved at each loading step data 共rich in terms of the variety of information, not just in terms
iteratively, simultaneously with the equilibrium iterations. Ožbolt of the number of specimens tested兲.
and Bažant 共1996兲 found that the modified nonlocal averaging A typical example is provided by the size effect method.
based on microcrack interactions makes it easier to reproduce, Sometimes it is assumed that the nonlocal characteristic length
with the same set of material parameters 共including the character- can be determined by fitting the size effect on the nominal
istic length兲, the results of different types of experiments, espe- strength of specimens, in other words, by trying to reproduce the
cially the failures dominated by tension and those dominated by dependence of the peak load on the structure size for a set of
shear and compression. geometrically similar fracture specimens covering a sufficiently

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1131

J. Eng. Mech., 2002, 128(11): 1119-1149

Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

Fig. 14. Left: cross section of concrete specimen restrained by elastic rods; right: load-displacement curves of restrained specimen and area W s

wide range of sizes 共or, more generally, brittleness numbers, in the size effect method. This again confirms that the nonlocal char-
the case of dissimilar specimens兲. In reality, this procedure can acteristic length cannot be uniquely determined from the size ef-
give only one material length parameter, which is related 共through fect alone.
a geometry factor determined from fracture mechanics兲 to the For a reliable identification of the nonlocal length parameter,
length of the process zone rather than to its width. one needs the results of tests that are sensitive to the nonlocal
A rigorous determination of the nonlocal characteristic length aspects of the material behavior. One possible solution is to com-
merely from the size effect on the peak loads for one test geom- bine two kinds of tests: 共1兲 those leading to localized damage
etry is, therefore, impossible. Only if one takes into account the patterns with 共2兲 those where the damage remains distributed. The
fact that both dimensions of the process zone are related to the former kind is sufficiently represented by standard fracture tests
same microstructural property, e.g., to the maximum aggregate on notched specimens. The latter kind is represented, for instance,
size and spacing in concrete, it is possible to assume that the ratio by the elastically restrained tensile test, proposed by Bažant and
between the two length parameters is equal to a certain constant Pijaudier-Cabot 共1989兲 and later modified by Mazars and Ber-
and relate the length provided by the size effect method to the thaud 共1989兲, and Mazars et al. 共1990兲. In this kind of experi-
nonlocal characteristic length. In practice, this assumption can be ment, tension is applied to a long prismatic specimen of concrete,
hidden in fixing certain other parameters, e.g., those that define restrained on the surface by gluing to it many thin longitudinal
the local softening law. aluminum rods 共Fig. 14兲. The cross section of the rods must be
For a gradient damage model, the optimal estimation of the sufficient to prevent the tangential stiffness of the specimen from
characteristic length of cement mortar reinforced by cellulose fi- ever being negative, despite softening damage in concrete. Fur-
bers was studied in detail by Carmeliet 共1999兲. His conclusions thermore, to permit simple analysis, the concrete thickness must
remain valid for nonlocal models of the integral type. Carmeliet be small enough 共only a few aggregate sizes兲 to force the defor-
used the Markov estimation method, which gives not only the mation to be essentially uniform across the thickness. The area of
mean values of the material parameters, but also their standard the load-displacement curve lying above the line that corresponds
deviations and correlation coefficients, provided that a sufficient to the stiffness of the elastic rods alone 共without the concrete兲
number of test results are available. The identification procedure represents the energy W s dissipated by strain-softening damage
exploited the results of uniaxial tensile tests on a double-notched within the specimen. Dividing W s by the volume V of tensioned
specimen, and of four-point bending tests on single-notched concrete in the specimen, one gets an estimate of the dissipation
specimens of different sizes in the range 1:8 and with two differ- density per unit volume, g F ⫽W s /V, which is valid in the absence
ent relative notch depths. Young’s modulus was determined di- of localization and should correspond to the area under the local
rectly from the tensile tests, and the Markov method aimed at stress–strain diagram. Once the parameters of the local stress–
determining the tensile strength, the nonlocal characteristic strain law are fixed, the nonlocal characteristic length remains the
length, and a parameter that appears in the damage law and con- only undefined parameter. It can be determined from the ratio
trols the area under the local stress–strain curve, i.e., the dissipa- G F /g F where G F is the fracture energy 共dissipation per unit area兲,
tion density per unit volume in the absence of localization. When which can be determined from fracture tests. In fact, it is not even
the identification was based on the peak loads from all the tests, necessary to use the size effect method; the average peak load for
Carmeliet found quite a high correlation between the strength and a single specimen size would suffice, provided that the specimen
the characteristic length 共correlation coefficient 0.88兲. This indi- is large enough to make the process zone substantially smaller
cates a strong dependence between these parameters, meaning than the whole specimen.
that the procedure just outlined cannot determine them It must, nevertheless, be noted that the aforementioned experi-
independently—they are somehow constrained by an implicit hy- mental procedure would be perfect only if the elastically re-
pothesis. Even if the identification procedure is refined by taking strained tensile test provided reliable information on the complete
into account the entire load-displacement curves 共not only the stress–strain law, including the value of the tensile strength. How-
peak loads兲, the correlation coefficient remains between 0.84 and ever, a refined analysis by Boudon-Cussac et al. 共1999兲 revealed
0.95, depending on which experiments are taken into account. that this technique can have a large error due to partial debonding
Correlation coefficient 0.95 is obtained when the parameters are at the interface between the concrete and the glued rods. If any
evaluated only from the geometrically similar specimens with the debonding occurs, the true behavior of concrete is not easy to
same relative notch size, which is the typical case considered by deduce, because the stress state in concrete ceases to be uniform,

1132 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 even in the central part of the specimen. Instead of a straightfor- calization due to softening and the deviations from locality in the
ward interpretation of the test results, one would have to resort to elastic range are not expected to play an important role.
inverse analysis. Another spoiling effect, albeit no doubt less se- In the inelastic range, the nonlocal constitutive law could be
rious, is that a surface restraint by homogeneously deforming rods defined as 共Bažant et al. 1996兲
does not exactly replicate the ideal microstructural deformation at
a cross section of a large concrete specimen with statistically ␴⫽De ⑀⫺s共 ⑀兲 (35)
homogeneous deformation. where s(⑀)⫽nonlocal average of the inelastic stress s evaluated
An alternative identification procedure can be based on local from the local strain ⑀. Unfortunately, this averaging scheme is
measurements in or around the process zone. Knowing the details suitable only at the early stages of softening but later inevitably
of the local strain distribution, one can evaluate the nonlocal char- leads to spurious locking effects. This has already been docu-
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

acteristic length by looking for the best match between the ex- mented in Fig. 5共b兲 for the case when the local constitutive law is
perimental and numerical strain profiles. Geers et al. 共1996a兲 ap- based on damage mechanics. In a general case, the reason for
plied this idea with success to short glass-fiber-reinforced locking can be explained as follows.
polypropylene, based on the experimental results reported by Under uniaxial tension with uniform strain, the actual stress
Geers et al. 共1996b兲, who measured the displacements at a num- ␴⫽E⑀⫺s is expected to be non-negative. Consequently, in the
ber of points arranged in a grid on the surface of the compact general case with nonhomogeneous strain, the inelastic stress
tension 共wedge splitting兲 specimen. For concrete, Denarié et al. evaluated from the local stress–strain law cannot exceed the elas-
共2001兲 tried to measure strains inside the process zone using op- tic stress, which can be written as s(x)⭐E⑀(x) for all x. Now, let
tical fibers with Bragg gratings, but the accuracy of this technique x max be the spatial coordinate of the point at which the strain
in the presence of cracking and the correct interpretation of the attains its maximum value ⑀ max . It is easy to show that the non-
results are not completely clear. Nevertheless, these experiments local inelastic stress at x max must satisfy the inequality

冕 冕
confirmed that the process zone has a finite width, because the
measured strains were way above the elastic limit, even though s̄ 共 x max兲 ⫽ ␣ 共 x max ,␰ 兲 s 共 ␰ 兲 d␰⭐ ␣ 共 x max ,␰ 兲 E⑀ 共 ␰ 兲 d␰
the fibers were not crossed by the final macroscopic crack. An- L L
other interesting point is that the residual strains measured at
complete failure were rather high. This confirms the fact, well
known from fracture testing of concrete, that the assumption of
⭐E⑀ max 冕 L
␣ 共 x max ,␰ 兲 d␰⫽E⑀ max (36)
unloading to the origin, characteristic of pure damage models, is
simplistic. Realistic models should incorporate permanent strains, The derivation exploits the basic properties of the nonlocal weight
even under tension. function, namely, the fact that ␣(x,␰) is non-negative and nor-
Although detailed information on the precise distribution of malized according to Eq. 共20兲. From the one-dimensional version
strains in the process zone is highly valuable, it is difficult to of Eq. 共35兲 it is clear that the stress at x max can vanish only if
obtain. For identification of the nonlocal characteristic length, it is E⑀(x max)⫽s̄(xmax), i.e., if Eq. 共36兲 holds with equality signs. But,
nevertheless sufficient to know at least the width of the process this is possible only if ⑀(␰)⫽⑀ max for all the points within the
zone. An estimate can be obtained by monitoring and evaluating domain of influence of point x max , i.e., for all ␰ such that
the acoustic emissions produced by cracking, or by x-ray inspec- ␣(x max ,␰)⬎0. Note that under uniaxial tension with negligible
tion. Such experimental techniques have been developed among body forces the stress is uniform along the entire bar. The fore-
others by Mihashi and Nomura 共1996兲, Landis 共1999兲, Otsuka and going analysis means that the stress in the bar can vanish only if
Date 共2000兲, or Labuz et al. 共2001兲. the strain distribution is uniform along the entire bar. In other
words, the strain distribution cannot remain localized until com-
plete failure. This is true independently of the type of constitutive
Nonlocal Adaptation of General Constitutive Models law, as long as this law does not generate compressive stress
under monotonic uniaxial tension.
The section on nonlocal damage models dealt with various non- Consequently, the nonlocal formulation with averaging of the
local formulations of the simple isotropic damage model with one inelastic stress is inherently incapable of describing complete fail-
scalar damage variable. It would be very useful to develop a ure, and so it cannot be used as a unifying concept. The same
unified nonlocal formulation applicable to any inelastic constitu- holds for the modifications that define the inelastic stress incre-
tive model with softening and acting as a reliable localization mentally or evaluate the inelastic stress from the nonlocal strain.
limiter, insensitive to mesh bias and free of locking effects. How-
ever, such a formulation is not available.
The first four formulations from Table 1 rely on the concept of Nonlocal Plasticity
damage and, consequently, they can be generalized only to aniso-
tropic damage models or to the closely related smeared crack
models. For other constitutive frameworks, such as plasticity or
Associated Plasticity with Isotropic Hardening or
microplane theory, a straightforward generalization is not avail-
Softening
able. In the standard 共local兲 version of the flow theory of plasticity with
The formulations based on the notion of inelastic stress may isotropic hardening or softening, the yield function, typically, has
seem to be ideal candidates for a unifying averaging scheme, the form
because the inelastic stress s can be defined for any type of con-
f 共 ␴,␬ 兲 ⫽F 共 ␴兲 ⫺␴ Y 共 ␬ 兲 (37)
stitutive model, simply as the difference between the elastically
evaluated stress, De ⑀, and the actual stress, ␴; see Fig. 12. In the where ␴⫽stress, ␬⫽scalar hardening-softening variable, F(␴)
elastic range, the inelastic stress by definition vanishes, and so the ⫽equivalent stress 共e.g., the von Mises equivalent stress for J 2
model response remains local. In the present context, this is a plasticity, or the maximum principal stress for Rankine plasticity兲,
desirable feature, because we are interested mainly in strain lo- and ␴ Y ⫽current yield stress. The evolution of the yield stress as

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1133

J. Eng. Mech., 2002, 128(11): 1119-1149

 a function of the hardening variable is described by the hardening Eringen 共1981兲 started from the plasticity theory formulated in
law. This law is, for future use, conveniently written as the strain space by Green and Naghdi 共1965兲 and Naghdi and
Trapp 共1975兲. The yield function is here written in terms of the
␴ Y 共 ␬ 兲 ⫽␴ 0 ⫹h 共 ␬ 兲 (38)
strain, plastic strain, and hardening variable共s兲. The original
where ␴ 0 ⫽initial yield stress and h⫽hardening function. The theory was developed for large-strain applications, but for the
derivative H⫽dh/d␬ is called the plastic modulus. The funda- present purpose we could write the yield condition as
mental equations of associated elastoplasticity include also the
g 共 ⑀,⑀p ,␬ 兲 ⫽0 (46)
elastic stress–strain law,
␴⫽De 共 ⑀⫺⑀p 兲 (39) This condition is equivalent to the classical yield condition in the
stress space, f (␴)⫽␴ Y (␬), if the function g is defined as
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

the flow rule,

g 共 ⑀,⑀p ,␬ 兲 ⫽F 关 De 共 ⑀⫺⑀p 兲兴 ⫺␴ Y 共 ␬ 兲 (47)
⑀˙ p ⫽␭˙ f共 ␴兲 (40)
The evolution laws for the internal variables and the loading–
and the loading–unloading conditions unloading conditions are postulated in the general form

␭˙ ⭓0, f 共 ␴,␬ 兲 ⭐0, ␭˙ f 共 ␴,␬ 兲 ⫽0 (41) ⑀˙ p ⫽␭˙ ␳共 ⑀,⑀p ,␬ 兲 (48)

Here, ⑀⫽(total) strain, ⑀p ⫽plastic strain, De ⫽elastic material ␬˙ ⫽␭˙ k 共 ⑀,⑀p ,␬ 兲 (49)
stiffness, ␭˙ ⫽rate of the plastic multiplier, and f⬅⳵ f /⳵␴
⫽gradient of the yield function, defining the direction of plastic g 共 ⑀,⑀p ,␬ 兲 ⭐0, ␭˙ ⭓0, ␭˙ g 共 ⑀,⑀p ,␬ 兲 ⫽0 (50)
flow.
The hardening variable ␬ reflects the changes in the micro- where ␳ and k are given functions. In analogy to the stress–space
structure induced by plastic flow. Its rate is usually related to the formulation, the rate of the plastic multiplier
plastic strain rate by the strain-hardening hypothesis
␬˙ ⫽ 储 ⑀˙ p 储 (42) 冓冉 冊 冔 ⳵g
⳵⑀
T
⑀˙

冉 冊
␭˙ ⫽⫺ (51)
or by the work-hardening hypothesis ⳵g T
⳵g
␳⫹ k
⳵⑀ p ⳵␬
␴T ⑀˙ p
␬˙ ⫽ . (43) can be evaluated from the consistency condition ġ⫽0, valid dur-
F 共 ␴兲
ing the plastic flow.
Substituting the plastic strain rate 共40兲 into Eq. 共42兲 or 共43兲, we In the nonlocal plasticity theory of Eringen 共1981兲, the stress is
can write computed by averaging the local stress that would be obtained
from the local model. This is equivalent to rewriting the stress–
␬˙ ⫽k␭˙ (44) strain law 共39兲 in the form
T
where k⫽ 储 f(␴) 储 for strain hardening and k⫽␴ f(␴)/F(␴) for
work hardening. ␴⫽De 共 ⑀⫺⑀p 兲 (52)
During plastic yielding, the yield function must remain equal If the elastic stiffness is constant in space, Eq. 共52兲 is equivalent
to zero, and so the rates of the basic variables must satisfy the to
consistency condition ḟ ⫽0. This makes it possible to derive the
well-known expression for the rate of the plastic multiplier, ␴⫽De 共 ⑀⫺⑀p 兲 (53)

具 fT De ⑀˙ 典 which is a straightforward extension of the nonlocal elastic law

␭˙ ⫽ T (45) 共12兲. Thus, the nonlocal approach of Eringen can be characterized
f De f⫹kH
as averaging of the stress or of the elastic strain. The response of
where the Macauley brackets 具¯典 denote the positive part; for the model is nonlocal already in the elastic range, which is not a
more details, see, e.g., Jirásek and Bažant 共2001兲, Sec. 20.1. desirable feature in applications to problems with strain localiza-
In the presence of softening, characterized by a negative value tion due to softening. Stress averaging does not act as a localiza-
of the plastic modulus H, the boundary value problem becomes ill tion limiter anyway. For instance, in the one-dimensional tensile
posed and must be regularized. This can be achieved by a suitable test, all the solutions obtained with the corresponding local model
nonlocal formulation. would remain admissible, since the stress is constant due to the
equilibrium condition and the constant field is not modified by the
averaging operator.
Eringen’s Nonlocal Plasticity
As a further step, Eringen 共1983兲 formulated nonlocal theories
Historically, the first nonlocal formulations of plasticity were pro- of plasticity in the stress space. In the flow theory he considered
posed by Eringen in the early 1980s. He set up the framework for only perfect von Mises plasticity with an associated flow rule, but
three classes of nonlocal plasticity models, based on the deforma- the logic he followed would, in a general case, lead to the stress–
tion theory, flow theory, and strain–space plasticity. Eringen did strain law
not consider the case of softening and did not intend his models to
serve as localization limiters. Rather, he was interested in the ␴⫽De ⑀⫺De ⑀p (54)
continuum-based description of interacting dislocations and in the which is in the case of a spatially constant elastic stiffness equiva-
nonlocal effects on the distribution of stress around the crack tip lent to
in elastoplastic fracture mechanics. Nevertheless, for the sake of
comparison it is useful to briefly describe his approach. ␴⫽De 共 ¯⑀⫺⑀p 兲 (55)

1134 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 Thus, one could say that the averaged quantity is the elastic stress
or the total strain. Again, the response is nonlocal already in the
elastic range, and the formulation cannot be used as a localization
limiter, but this time the reason is different. While the previously
mentioned model with nonlocal stress would not prevent localiza-
tion into a set of zero measure, the present model with nonlocal
strain would not allow any localization at all 共in the one-
dimensional test problem兲. This becomes clear if one realizes that,
at the bifurcation from a uniform strain state, the nonlocal strain
rate in the elastically unloading region would have to be equal to
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

a negative constant while in the plastically softening region it

would have to be equal to a positive constant. However, the non-
local strain obtained by applying an integral operator with a con-
tinuous weight function is always continuous, even in situations
when the local strain has the character of a Dirac distribution.
Consequently, the nonlocal strain cannot have a jump at the elas-
toplastic boundary, and the plastic strain cannot localize into any
region smaller than the entire bar.

Nonlocal Plasticity Model of Bažant and Lin

The first nonlocal formulation of softening plasticity was pro-
posed by Bažant and Lin 共1988b兲; it was applied in finite-element
analysis of the stability of unlined excavation of a subway tunnel
in a grouted soil. The underlying cohesive–frictional plasticity
model was based on the Mohr–Coulomb yield condition and a
linear softening law with the work-softening hypothesis. Bažant
and Lin 共1988b兲 proposed to replace the plastic strain in the
stress–strain law 共39兲 by its nonlocal average,

⑀p 共 x 兲 ⫽ 冕
V
␣ 共 x,␰兲 ⑀p 共 ␰兲 d␰ (56)

where the local plastic strain ⑀p is obtained by integrating in time

the rate ⑀˙ p evaluated from the standard expressions 共45兲 and 共40兲.
As an alternative, Bažant and Lin 共1988b兲 also proposed to aver-
age the rate of the plastic multiplier 共45兲 and then substitute the
nonlocal average

¯␭˙ 共 x兲 ⫽
冕
V
␣ 共 x,␰兲 ␭˙ 共 ␰兲 d␰ (57)

into Eq. 共40兲 instead of ␭˙ . They noted that this modification is

computationally more efficient 共since the averaged quantity is a
scalar and not a tensor兲 and that it leads to very similar numerical Fig. 15. Nonlocal plastic analysis of subway tunnel excavation in a
results as the formulation with averaging of the plastic strain. grouted soil 共after Bažant and Lin 1988b兲: 共a兲 finite-element meshes,
Fig. 15共a兲 shows four successive mesh refinements used in the 共b兲 deformed mesh with N⫽608 degrees of freedom, and 共c兲 contours
tunnel excavation analysis, and the 共exaggerated兲 deformation of of the strain-softening zone.
the mesh caused by the excavation 共Bažant and Lin 1988b兲 is
plotted in Fig. 15共b兲. The contours of the strain-softening zone provided that the elastic stiffness De is constant in space. Conse-
given in Fig. 15共c兲 demonstrate negligible differences, which con- quently, this nonlocal plasticity model is expected to exhibit lock-
firm that averaging of the plastic strain or of the plastic multiplier ing at later stages of the process of strain softening.
indeed acts as a localization limiter. However, this particular
model with
Plasticity with Nonlocal Softening Variable
␴⫽De 共 ⑀⫺⑀p 兲 (58) Perhaps the simplest nonlocal formulation of plasticity with iso-
turns out to be a special case of the nonlocal formulation with tropic strain softening is obtained if the current yield stress is
averaging of the inelastic stress, already discussed in the section computed from the nonlocal average of the softening variable.
on nonlocal adaptation of general constitutive models. In local The softening law 共38兲 is replaced by
plasticity, the expression for the inelastic stress reads ␴ Y ⫽␴ 0 ⫹h 共 ¯␬ 兲 (61)
s⫽De ⑀⫺␴⫽De ⑀⫺ 关 De 共 ⑀⫺⑀p 兲兴 ⫽De ⑀p (59) where

冕
and the corresponding nonlocal inelastic stress is
¯␬ 共 x兲 ⫽ ␣ 共 x,␰兲 ␬ 共 ␰兲 d␰ (62)
s̄⫽De ⑀p ⫽De ⑀p (60) V

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1135

J. Eng. Mech., 2002, 128(11): 1119-1149

 ized by a high degree of regularity. This is the case, for instance,
for nonlocal damage and smeared crack models. For nonlocal
softening plasticity, regularity can be achieved by a modification
of the variable driving the yield stress degradation. Vermeer and
Brinkgreve 共1994兲 defined the variable ¯␬ that enters the softening
law 共61兲 as a linear combination of the local cumulative plastic
strain and its nonlocal average:

Fig. 16. Geometrical meaning of dissipation length ᐉ d

¯␬ 共 x兲 ⫽ 共 1⫺m 兲 ␬ 共 x兲 ⫹m 冕
V
␣ 共 x,␰兲 ␬ 共 ␰兲 d␰ (64)
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

The same definition of the softening variable was also considered

by Planas et al. 共1996兲 and by Strömberg and Ristinmaa 共1996兲.
is the nonlocal softening variable. The flow rule and the evolution Note that Eq. 共62兲 is obtained as a special case with m⫽1.
law for the local softening variable ␬ are kept unchanged. Another special case with m⫽0 corresponds to the local soften-
As shown by Planas et al. 共1993兲, this simple nonlocal plastic- ing plasticity model. One might expect the typical values of m
ity model is, in fact, equivalent to a cohesive crack model 共Hill- used by the generalized model to lie between these two cases but,
erborg et al. 1976兲. The plastic strain localizes into a set of zero interestingly, the plastic region has a nonzero width only for m
measure but, in contrast to the local model, the energy dissipation ⬎1 共which may be called an overnonlocal formulation兲.
does not vanish. The total work G F spent by the complete failure The localized strain distributions in a one-dimensional tensile
共material separation兲 over a unit area of the cohesive crack under test obtained with various values of m are plotted in Fig. 17. Far
uniaxial tension is equal to the product g F ᐉ d , where g F is the from the boundary, the distribution of plastic strain inside the
area under the local stress–strain curve 共work per unit volume in plastic region is symmetric 关Fig. 17共a兲兴. If the plastic region is
a local continuum兲 and l d is the equivalent dissipation length adjacent to the physical boundary of the bar, the largest plastic
imposed by the nonlocal model. It can be shown that, for the strain is attained right on the boundary 关Fig. 17共b兲兴. The width of
present model, the plastic region is an increasing function of parameter m and
1 tends to zero for m→1 ⫹ . For a given m, the width of the plastic
ᐉ d⫽ (63) region is proportional to the internal length imposed by the non-
␣ 共 xs ,xs 兲
local averaging function. It is worth noting that the shape of the
where xs ⫽point at which the localized cohesive crack forms. Far plastic strain profile does not depend on the softening modulus,
from the boundary, ᐉ d ⫽1/␣ ⬁ (0) is a constant proportional to the and that the width of the plastic region does not change during the
internal length parameter of the nonlocal weight function. It has softening process 共in the uniaxial test兲.
the geometrical meaning of the width of a rectangle that has the Since the yield stress in Eq. 共61兲 depends on the nonlocal
same area as the nonlocal weight function and the same height at softening variable, the yield function and the loading–unloading
the origin, see Fig. 16. In the proximity of a physical boundary, conditions have a nonlocal character, and the stress evaluation
the dissipation length 共63兲 decreases, which means that the local cannot be performed at each Gauss integration point separately.
fracture energy is smaller than in an infinite body. It has been This is the price to pay for the enhancement of the standard con-
suggested 共Hu and Wittmann 2000兲 that this could partially ex- tinuum description, which restores the well posedness of the
plain the size effect on the fracture energy observed in experi- boundary value problem and leads to numerical solutions that
ments with concrete and mortar. exhibit no pathological sensitivity to finite-element discretization.

Formulation Combining Nonlocal and Local Softening Thermodynamics-Based Symmetric Nonlocal Plasticity
Variables
One noteworthy aspect of the nonlocal models presented so far
The singular character of the strain distribution generated by the has not yet been mentioned—the fact that they yield nonsymmet-
simple nonlocal plasticity model from the previous section is an ric tangential stiffness matrices. This point was studied, and the
exception. As a rule, nonlocal models lead to solutions character- cause explained for the nonlocal damage model 共Bažant and

Fig. 17. Plasticity model combining local and nonlocal softening variables: Effect of parameter m on the plastic strain profiles localized 共a兲 inside
the bar, and 共b兲 at the boundary.

1136 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 Pijaudier-Cabot 1988兲. The structure of the nonlocal stiffness ma-
trix was described by Huerta and Pijaudier-Cabot 共1994兲, and the 冕V
f 共 x兲 ḡ 共 x兲 dx⫽ 冕 冕 V
f 共 x兲
V
␣ 共 x,␰兲 g 共 ␰兲 d␰dx
numerical implementation and convergence rates were studied by
Jirásek and Patzák 共2002兲. No convergence problems attributed to
nonsymmetry have so far been encountered in computations with
the aforementioned models.
⫽ 冕冕 V V
␣ 共 x,␰兲 f 共 x兲 dx g 共 ␰兲 d␰ (68)

Modifications achieving symmetry would nevertheless be we obtain f̃ (␰)⫽ 兰 V ␣(x,␰) f (x) dx, or, equivalently,
highly desirable, not only for theoretical reasons but also for im-
proving computational efficiency of implicit solution schemes
共through the use of more efficient equation solvers兲. A symmetric
f̃ 共 x兲 ⫽ 冕 ␣ 共 ␰,x兲 f 共 ␰兲 d␰ (69)
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

V
structure of nonlocal models can be obtained by an appropriate
generalization of the postulate of maximum plastic dissipation This formula indicates that the dual averaging operator differs
共understood here in the global sense兲. The thermodynamic frame- from the original 共primal兲 one only in the order of arguments of
work established for nonlocal elasticity by Edelen and Laws the weight function ␣. For an infinite domain V ⬁ , the weight
共1971兲 and Edelen et al. 共1971兲 served as a basis for thermody- function is symmetric 共because it depends only on 储 x⫺␰储 ), and
namically consistent formulations of nonlocal softening plasticity so there is no difference between the primal and dual averaging
共Svedberg 1996; Svedberg and Runesson 1998; Polizzotto et al. 共thus the nonlocal operator is self-adjoint兲. For a finite domain V,
1998兲. the weight function is nonsymmetric in the proximity of the
From the thermodynamic point of view, the elastic stress– boundary, and f̃ is, in general, different from f̄ . In particular, the
strain law and the plastic hardening law are state laws that can be dual averaging operator does not satisfy the normalizing condi-
derived from a suitably defined free-energy potential. For models tion that a constant field f is transformed into a constant field f̃
formulated within the framework of generalized standard materi- ⬅f.
als 共complying with the postulate of maximum plastic dissipa- As pointed out by Rolshoven and Jirásek 共2001兲, the nonlocal
tion兲, the flow rule, the definition of the hardening variable and plasticity formulation with softening law 共66兲 does not provide
the loading–unloading conditions are complementary laws that full regularization, similar to the simpler formulation with soften-
can be derived from the dual dissipation potential defined as the ing law 共61兲. Plastic strain still localizes into a set of zero mea-
indicator function of the set of plastically admissible states; for a sure. This is not possible in the nonlocal plasticity theory pro-
detailed discussion, see, e.g., Jirásek and Bažant 共2001, Chap. posed by Svedberg and Runesson 共1998兲, who considered the free
23兲. The postulate of maximum plastic dissipation restricts the energy as a function of both the local and nonlocal internal vari-
class of models covered by this framework to those that satisfy ables. A particular model of this type defines the plastic part of the
the conditions of convexity 共of the yield function兲 and normality free energy in the form ␺ p (␬,␬ ¯ )⫽␺ p1 (␬)⫹␺ p2 (␬¯ ⫺␬), i.e., as a
共of the plastic flow and of the evolution of hardening variables兲. sum of two terms that depend on the local softening variable and
on the difference between the nonlocal and local softening vari-
For instance, models with nonassociated flow rules remain out-
able, respectively. The consistent thermodynamic approach then
side the class of generalized standard materials, even though they
leads to the hardening–softening law
can still be thermodynamically admissible 共Bažant and Cedolin
1991兲. Nevertheless, the postulate of maximum dissipation leads ␴ Y ⫽␴ 0 ⫹h 1 共 ␬ 兲 ⫹hg ␬ ⫺␬ 兲 ⫺h 2 共 ¯␬ ⫺␬ 兲
2共 ¯ (70)
to a symmetric structure of the constitutive equations, and it is
interesting to explore its possible extensions to nonlocal material where function h 1 ⬅␳␺ ⬘p1 describes softening under homogeneous
models. conditions 共uniform distribution of plastic strain兲 and function
For plasticity with isotropic softening, Borino et al. 共1999兲 h 2 ⬅␳␺ ⬘p2 is the correction due to nonlocal effects. This model
considered the free-energy density in the form has been shown to give a localized plastic zone of a nonzero size
共Borino and Failla 2000兲. However, its scope of application is
␳␺ 共 ⑀,⑀p ,␬
¯ 兲 ⫽␳␺ e 共 ⑀⫺⑀p 兲 ⫹␳␺ p 共 ¯␬ 兲 (65) limited; while it reasonably captures the initial localization pro-
file, it cannot describe complete failure with zero residual resis-
where ␳␺ e (⑀⫺⑀p )⫽ 21 (⑀⫺⑀p ) T De (⑀⫺⑀p )⫽elastically stored en- tance of the material. This can be easily explained for the uniaxial
ergy and ␳␺ p (␬
¯ )⫽plastic part of the free energy, usually inter- tensile test. If the material completely loses cohesion at late stages
preted as the energy stored in microstructural changes. The pos- of the deformation process, the local softening law must be such
tulate of maximum plastic dissipation then leads to an associated that ␴ 0 ⫹h 1 (␬)⫽0 for ␬ exceeding a critical value ␬ c . In the
nonlocal plasticity model with the softening law 共38兲 replaced by zone where the plastic strain exceeds ␬ c , the actual yield stress
共70兲 can vanish only if the nonlocal correction vanishes, i.e., if
␴ Y ⫽␴ 0 ⫹hg
共 ¯␬ 兲 (66)
hg ␬ ⫺␬ 兲 ⫽h 2 共 ¯␬ ⫺␬ 兲
2共 ¯ (71)
where ¯␬ ⫽(usual) nonlocal average of the softening variable ␬
共cumulative plastic strain兲, and the tilde over h(␬
¯ ) means that the Sufficiently far from the boundary, the dual averaging operator is
thermodynamic force q⫽␳⳵␺/⳵␬ ¯ ⫽␳d␺ p /d␬¯ ⫽h(␬ ¯ ) is subjected identical with the primal one. Arguments similar to those used by
to the so-called dual averaging. The dual averaging operator is Jirásek and Rolshoven 共2002兲 in their analysis of the Vermeer–
defined implicitly by the identity Brinkgreve nonlocal plasticity model 共previous section兲 lead to
the conclusion that Eq. 共71兲 is satisfied only if h 2 (␬
¯ ⫺␬) is con-
冕V
f̃ gdV⫽ 冕
V
f ḡdV (67)
stant, and this is possible only if ¯␬ ⫺␬ is constant, provided that
the hardening function h 2 is monotonic. Applying the same rea-
soning once again, it can be shown that the difference ¯␬ ⫺␬ is
which must hold for all functions f and g for which the right-hand constant across the plastic zone of a nonzero size only if ␬ is
side makes sense. Since constant along the entire bar. This means that when the plastic

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1137

J. Eng. Mech., 2002, 128(11): 1119-1149

 strain at the center of the localized plastic zone reaches ␬ c , the modynamically consistent. Moreover, the plastic strain again lo-
plastic zone starts expanding and the strength degradation gets calizes into a set of zero measure, similar to the model with a
delayed, which is perceived as a stress locking effect. For nonlin- nonlocal softening variable from the section on plasticity with
ear softening laws, this effect builds up gradually. nonlocal softening variable.
The foregoing thermodynamically based nonlocal plasticity
has apparently not yet been applied in multidimensional finite-
element codes. The same general idea was adapted to nonlocal Nonlocal Version of Dislocation-Based Gradient
damage by Benvenuti et al. 共2000兲. Comi and Perego 共2001兲 pro- Plasticity for Micrometer Scale
posed a simpler formulation that permits an explicit evaluation of While some numerical experts assert that the gradient models are
nonlocal averages, without the need for solving a nonlocal con- more friendly to a computer programer than the nonlocal averag-
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

sistency condition, which has the character of an integral equa- ing models, Gao and Huang 共2001兲 recently took the opposite
tion. However, since they used a primal averaging operator based view. For the purpose of finite-element analysis, they converted
on the scaled weight function 共21兲, the dual operator did not pre- the dislocation-based gradient plasticity model for metals on the
serve a uniform field, which caused certain problems in the vicin- micrometer scale 共Gao et al. 1999; Huang et al. 2000兲 to a form
ity of physical boundaries. A remedy was found by Borino et al. in which the strain gradient is approximated by a nonlocal inte-
共2002兲, who constructed a self-adjoint operator that preserves a gral over a representative volume. A generalized form of their
uniform field. The resulting thermodynamically based nonlocal idea is as follows.
damage model seems to perform well, but it still requires a two- Within a certain representative volume V c surrounding a given
fold nonlocal averaging, and so it is more computationally expen- point x, the strain tensor may be approximated as ⑀(x⫹s)
sive than the simple nonlocal damage theory of Pijaudier-Cabot ⬇⑀(x)⫹⑀,m (x)s m , where the subscript preceded by a comma de-
and Bažant 共1987兲. notes a partial derivative and repeated indices imply summation.
The thermodynamically based formulation is very attractive Multiplying this equation by s k and integrating over V c , one gets
from the theoretical viewpoint, and the fact that it leads to sym-
metric stiffness matrices is an asset from the computational view-
point. The dissipation inequality is automatically satisfied. How- 冕 Vc
⑀共 x⫹s兲 s k ds⬇⑀共 x兲 冕Vc
s k ds⫹⑀,m 共 x兲 冕
Vc
s m s k ds (74)
ever, there are two penalties to pay: 共1兲 a greater complexity of
the constitutive model; and 共2兲 a broadened bandwidth in the The integrals on the right-hand side characterize the geometry of
stiffness matrix, caused by the double application of the nonlocal the representative cell. If point x is located at the center of gravity
operator. Furthermore, while a very appealing theoretical frame- of the cell, the first-order moments 兰 V c s k ds vanish. The second-
work has been developed, much of its physical foundation re- order moments

冕
mains clouded.
The way to symmetrize the nonlocal model based on crack I mk ⫽ s m s k ds (75)
interactions is not known at present. It is even unclear whether an Vc
effort in that direction would be physically justified. are components of the inertia tensor, which is always regular, and
thus invertible. In the simplest case of a cubic or spherical cell,
Nilsson’s Model I mk is a multiple of the unit tensor. Multiplying Eq. 共74兲 from the
right by the inverse tensor I ⫺1
kl , we obtain for the strain gradient
In his thesis 共Nilsson 1994兲 and journal article 共Nilsson 1997兲, the approximation

冕
Nilsson proposed a thermodynamically motivated nonlocal plas-
ticity model that shares many similar features with the model of ⫺1
⑀,l 共 x兲 ⬇Ikl ⑀共 x⫹s兲 s k ds (76)
Borino et al. 共1999兲. However, there are also some important dif- Vc
ferences. In the most general version of his model, Nilsson con- If the strain gradient terms in a gradient plasticity model are re-
sidered all the arguments of the free-energy potential as nonlocal. placed by this approximation, the numerical solution can be pro-
Since this would lead to a stress–strain law that is not easily gramed in the same manner as the nonlocal averaging models.
invertible, Nilsson focused his attention on what he called the In strain-gradient plasticity, the size of the representative cell
model with restricted nonlocality, which deals with total strain is dictated by the characteristic length
only in its local form. For isotropic softening, the free-energy
density is written as ᐉ g ⫽␣ g 共 G/␴ 0 兲 2 b (77)

␺ 共 ⑀,⑀ ¯ 兲 ⫽␺ e 共 ⑀⫺⑀
¯ p ,␬ ¯ p 兲 ⫹␺ p 共 ¯␬ 兲 (72) where G⫽elastic shear modulus, ␴ 0 ⫽yield stress, b⫽Burgers
vector of edge dislocation, and ␣ g ⫽dimensionless semiempirical
In contrast to Eq. 共65兲, the plastic strain that appears in the elastic constant of the order of 1 共Gao et al. 1999兲. Alternatively, one
part of the free energy is nonlocal. Consequently, the elastic part could consider the approximation of the gradient by an integral as
of the stress–strain law 共39兲 is now replaced by a purely numerical procedure and select V c as the smallest neigh-
␴⫽De 共 ⑀⫺⑀
¯p兲 (73) borhood of the given point x that contains a sufficient number of
material points traced by the finite-element program. Eq. 共76兲 can
The softening law 共38兲 is the same as for the simple formulation then be interpreted as a generalized finite-difference scheme. In
of nonlocal plasticity, and the flow rule 共40兲 and loading– one dimension, the minimum required number of integration
unloading conditions 共41兲 remain standard. However, this is not points would be 2, and if these points are located symmetrically
consistent with the postulate of maximum plastic dissipation, as with respect to x, Eq. 共76兲 yields the standard finite-difference
pointed out by Borino and Polizzotto 共1999兲 and admitted by formula
Nilsson 共1999兲. It even turns out that, under certain circum-
stances, the global dissipation could become negative 共Jirásek and ⑀ 共 x⫹h 兲 ⫺⑀ 共 x⫺h 兲
⑀ ⬘共 x 兲 ⬇ (78)
Rolshoven 2002兲, which means that the model is not really ther- 2h

1138 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

Fig. 18. Adjustment of the stress–strain diagram according to the crack band approach: 共a兲 master curve and decomposition of strain into the
prelocalization and postlocalization parts, 共b兲 horizontal scaling of the postlocalization part for an element twice as large as the physical process
zone, 共c兲 snap back occurring for a very large element, and 共d兲 conservation of fracture energy by strength reduction.

However, in a general case 共with more integration points or ir- mesh is refined. This is why the crack band model 共or fracture
regular point arrangement兲, this approach does not provide the energy approach兲 cannot be considered as a true localization lim-
optimal accuracy, and it would be preferable to replace s k in Eq. iter. It provides only a partial regularization of the problem, in the
共74兲 by a suitably modified weight function. sense that the global response characteristics do not exhibit spu-
rious mesh sensitivity. However, the mesh-induced directional
bias is still present.
Partial Regularization and Substitutes for The crack band approach is based on the observation that, if a
Nonlocality local softening model is used, the numerically resolved process
zone typically localizes into one layer of finite elements, called
the crack band. The effective width of the crack band, h e f , can be
Crack Band Model
estimated using the rules proposed by Bažant 共1985b兲 and refined
The crack band model 共Bažant 1982; Bažant and Oh 1983兲 is by Rots 共1988兲. A reasonable estimate is obtained by projecting
共aside from viscosity兲 practically the simplest, and computation- one element onto the direction normal to the band, which is 共for
ally most effective, way to avoid the pathological sensitivity to quasibrittle failure兲 usually close to the direction of maximum
mesh refinement. Since the beginning, it has been the most widely principal strain. The user specifies the stress–strain relation for
used model for distributed cracking of concrete and geomaterials the basic case when h e f is equal to the actual effective width of
and has been incorporated in a number of commercial finite- the physical fracture process zone, L s , which is proportional to
element codes. The basic idea of the crack band model is very the intrinsic material length ᐉ. As far as possible, the size of the
similar to the mesh-adjusted softening modulus technique, pro- finite elements in the strain-softening distributed damage zone is
posed in the context of softening plastic shear bands by Pietrusz- kept equal to L s . But if, for reasons of computational efficiency,
czak and Mróz 共1981兲. Essentially, the same approach was devel- the finite elements need to be larger, then the postlocalization
oped by Willam et al. 共1986兲 under the name of the composite portion of the master constitutive diagram 共on which the tensorial
fracture model. constitutive law is based兲 is scaled horizontally 共i.e., in the direc-
This simple model is endowed with some, but not all, of the tion of the strain axis兲 by the factor L s /h e f ; see Fig. 18共b兲. This
characteristics of nonlocal models. It ensures the correct energy makes the strain softening steeper and a proper scaling ratio
dissipation in a localized damage band 共or equivalent fracture兲, achieves that the energy dissipated by the cracking band per its
and it gives the correct transitional size effect. The main differ- unit advance remains the same. Note that the postlocalization por-
ence from the nonlocal models is that the nonlocal spatial aver- tion is the difference, for the same stress, of the strain at the
aging is replaced by an energy-based rescaling of the postlocal- softening curve from the strain at the unloading diagram emanat-
ization part of the stress–strain relation, which takes into account ing from the point at which localization occurs 共typically the
the density of the computational mesh 共size of the finite ele- peak兲, not from a vertical line dropping down from that point
ments兲. The global load–displacement diagram can be captured 关Fig. 18共a兲兴. In the simple case of a triangular stress–strain dia-
correctly, but the width of the numerically resolved fracture pro- gram with linear strain softening in uniaxial tension characterized
cess zone depends on the element size and tends to zero as the by the tangent modulus E t ⬍0, the adjusted postpeak tangent

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1139

J. Eng. Mech., 2002, 128(11): 1119-1149

 modulus Ẽ t must satisfy the condition 共Bažant 1982; Bažant and should solve the problem with meshes of different inclinations.
The most reasonable solution is usually that providing the small-
Oh 1983兲
est peak load or the steepest postpeak load deflection diagram.
1 2G f 1 A caveat needs to be mentioned with respect to those infre-
⫺ ⫽ ⫺ (79) quent situations where the strain-softening damage does not lo-
Ẽ t h e f f t2 E
calize, being stabilized, for example, by a heavy enough rein-
where E⫽initial Young’s modulus and G f ⫽L s (E ⫺1 ⫺E ⫺1 2
t ) f t /2.
forcement net or by an adjacent layer of compressed material 共see
The increase of finite-element size, however, has a limit. For a the examples mentioned later in the section on limitations of co-
certain large enough critical element size, h crit , the adjusted post- hesive models兲. In such situations, the postpeak strain-softening
peak diagram develops a snapback 关Fig. 18共c兲兴, which is inadmis- behavior must not be rescaled. So, in using the crack band ap-
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

sible 共the stress would cease to be a unique functional of the proach, the user must separately assess whether it is reasonable to
strain history兲. To avoid the spurious snapback, one can use a expect localization. However, in some cases, diffuse softening
vertical stress drop and scale down the strength limit f t to some damage patterns in some parts of the structure can coexist with
equivalent value f eq such that h e f times the area under the stress– localized cracks in other parts, and these parts may even change
strain diagram with the vertical drop would remain equal to G f during the loading process. In such cases it is then next to impos-
关Fig. 18共d兲兴. It turns out that, generally, f eq ⬀1/冑h e f , and for the sible to define a reasonable rule for the adjustment of the stress–
special case of a triangular stress–strain diagram strain diagram according to the element size. A consistent solution
can be obtained only when finite-element sizes h⫽L s can be used
f eq ⫽ f t 冑2l 0 /h e f (80) 共in which case no rescaling is needed for the crack band ap-
where l 0 ⫽EG f / f 2t ⫽Irwin-typecharacteristic length of the mate- proach兲, or with a fully regularized nonlocal model.
rial. This expedient was proposed by Bažant and Cedolin 共1979兲 As explained before, the crack band model uses a modification
and Cedolin and Bažant 共1980兲, and good practical results with of the postpeak part of the constitutive law enforcing the correct
finite elements much larger than L s were obtained in situations energy dissipation by a localized crack band. When large finite
where failure is driven by fracture. elements need to be used 共larger than the actual width of the
To model a very narrow cracking band, or to approximate a fracture process zone兲, the resolution of the localization process
line fracture, finite elements smaller than L s may be used in the can be improved by special enrichments of the finite-element ap-
crack band model. The same scaling of the postpeak diagram can proximation that allow capturing narrow bands of highly local-
again be applied, making the strain softening less steep. As the ized strain inside the elements. The pioneering paper by Ortiz
mesh is refined, the band of cracking elements converges to a et al. 共1987兲 paved the way to special elements with embedded
layer of zero thickness, i.e., to a line 共in two dimensions兲 or localization bands 共Belytschko et al. 1988; Sluys 1997; Oliver
surface 共in three dimensions兲. Thus, the limit corresponds to the et al. 1998兲. These elements can describe a layer of softening
solution of a cohesive crack model. The mathematical aspects of material separated from the surrounding, elastically unloading
this correspondence were studied by Simo et al. 共1993兲. material by two parallel weak discontinuity planes across which
Scaling of the stress–strain diagram in the above sense is the displacement remains continuous but the out-of-plane strain
straightforward only for models that explicitly control the evolu- components can have a jump. The width of the band can be speci-
tion of inelastic strain, e.g., for softening plasticity or smeared fied independently of the mesh 共provided that the elements are not
crack models. In that case, the desired scaling effect is achieved too small兲, and it can be set equal to the actual width of the
by a modification of the hardening modulus 共derivative of stress process zone, L s , considered as a material parameter. A one-
with respect to inelastic strain兲. In continuum damage mechanics, dimensional element of this kind would give exactly the same
nonlinearity and softening are controled by the damage evolution response 共in terms of the relationship between the nodal displace-
law, and the reduction factor 1⫺␻ multiplies the total strain. It is, ments and nodal forces兲 as a standard element with the softening
therefore, not easy to scale only the postlocalization part of strain modulus adjusted according to the element size, as described for
while keeping the unloading part unaffected. An exact scaling the crack band model. For two- and three-dimensional elements,
procedure leads to a nonlinear equation 共except for the special the formulation with embedded discontinuities can better repro-
case of linear softening兲, which must be solved iteratively during duce the kinematics of highly localized deformation modes and
each stress evaluation. It is also difficult to extend the crack band thus avoid certain locking effects 共Jirásek 2000a兲.
approach to mixed-mode failure or failure under complex three- In a recent study 共Bažant et al. 2001; Červenka et al. 2002兲, a
dimensional stress states. In such general cases, the standard frac- new model for a damage localization layer within a finite element
ture energy is not sufficient to characterize the dissipation in lo- has been developed. For the special case of an orthogonal brick
calized failure modes. Generalized forms of the crack band element 共whose nodal displacements are compatible with uniform
approach that cover a variety of failure modes have been pro- strain兲, the model reduces to a combination of one brick element
posed, e.g., for softening plasticity models 共Etse and Willam for the strain-softening localization band and another brick ele-
1994; Kang 1997兲 and for the microplane constitutive model M4 ment for the parallel layer of elastically unloading material, such
for concrete 共paper in preparation by Bažant, Zi, Jendele, and that the interface conditions of stress equilibrium and of kine-
Novák兲. matic compatibility are exactly satisfied. In the general case of an
Whereas the nonlocal models with at least three finite elements arbitrary finite element containing a localization layer of arbitrary
across the width of the localization band are the best known way orientation, the incremental stress–strain relation is assumed to be
to avoid mesh orientation bias for the propagation direction 共in the same as in an ‘‘equivalent’’ orthogonal brick element that has
fact, better than the known mesh-adaptive schemes兲, the crack the same total volume, and the same volume and width of the
band model is poor in this regard. It performs best if the path of localization layer. The interface conditions 共numbering 6兲 are
the cracking band 共or the fracture to be approximated by it兲 is solved within each loading step or iteration, together with itera-
known in advance and if the mesh is laid out so that a mesh line tions for the nonlinear constitutive law. Implementation of this
would coincide with this path. If the crack path is not known, one localization element together with microplane model M4 for con-

1140 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 crete 共Bažant et al. 2000兲 led to good results for various localiza- linear elastic model with a path-independent work of separation.
tion problems. Such a simple approach is suitable only if the crack opening
grows monotonically, but it fails to give reasonable results for
Cohesive Crack and Cohesive Zone Models 共even partially兲 closing cracks. Moreover, the model has a non-
zero initial compliance, which means that the displacement jump
Traction-Separation Law starts growing whenever nonzero tractions are applied, no matter
The cohesive crack model for concrete, formulated by Hillerborg how small they are. This is appropriate for simulations of preex-
et al. 共1976兲 under the name of fictitious crack model, represents isting material interfaces 共grain boundaries, matrix-inclusion in-
a refinement of the classical cohesive models, with the main dif- terfaces, etc.兲 with a well-defined geometrical structure that can
ference that a cohesive crack is not considered to initiate and be taken into account by the finite-element mesh. When the aim is
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

propagate along a predetermined path but is assumed to initiate to simulate a crack propagating along an arbitrary path that is not
anywhere in the structure if the tensile stress reaches the strength known in advance, potential surfaces 共or lines, in two dimensions兲
limit. Although this approach is not really a nonlocal model, it of decohesion must be interspersed throughout the material 共Xu
provides an alternative to achieve one objective of the nonlocal and Needleman 1994兲. Their high initial stiffness has an adverse
models—a mesh-size independent response in the presence of effect on the conditioning of the global stiffness for implicit meth-
strain-softening damage or fracture. The early applications in the ods, or on the critical time step for explicit methods. Moreover,
field of concrete fracture were limited to mode-I situations. An due to the finite number of potential discontinuity segments, the
extension to cracks and interfaces opening under mixed-mode crack propagation path is locally always constrained to a discrete
conditions was developed, e.g., by Červenka 共1994兲, who ex- set of directions, typically spaced by 45 or 60°.
ploited a loading function in the space of normal and shear trac- A cohesive zone model with a nonzero initial compliance was
tions, originally proposed by Carol and Prat 共1990兲 in the context also used by Tvergaard and Hutchinson 共1993兲 and Wei and
of statically constrained microplane models. Scaling properties Hutchinson 共1999兲, who studied the interplay between plastic
and asymptotic solutions of cohesive crack models were studied yielding in a small process zone and separation processes at an
by Planas and Elices 共1992, 1993兲 and Bažant and Li 共1995兲. interface between two materials.
Phenomena such as debonding, delamination, or intergranular Ortiz 共1988兲 suggested a derivation of the traction-separation
damage in composites and metals are often described by cohesive law for mode I from a micromechanical model based on an array
zone models, going back to the ideas of Dugdale 共1960兲 and of collinear microcracks. Camacho and Ortiz 共1996兲 studied frac-
Barenblatt 共1962兲 and recently developed in a modern computa- ture and fragmentation in brittle materials using a cohesive model
tional framework by Needleman 共1987兲, Tvergaard and Hutchin- with possible crack initiation under mixed mode. The model was
son 共1992兲, Tvergaard and Hutchinson 共1993兲, and Ortiz and co- then reformulated within a thermodynamic framework 共Pandolfi
workers 共Camacho and Ortiz 1996; Pandolfi et al. 1999; Ortiz and et al. 1999兲 and extended to the area of finite opening displace-
Pandolfi 1999兲. ments 共Ortiz and Pandolfi 1999兲. Due to insistence on a simple
Cohesive models describe highly localized inelastic processes symmetric formulation with a potential, the resultant of the shear
by traction-separation laws that link the cohesive traction trans- and normal stresses in the cohesive crack has in these models the
mitted by a discontinuity line or surface to the displacement jump, same direction as the relative 共normal and shear兲 displacement
characterized by the separation vector. The physical interpretation across the crack. For shear with a nearly vanishing normal stress,
and the numerical implementation of such models can be mani- the relative displacement is predicted to be nearly parallel to the
fold. The discontinuity surface can be an internal boundary be- crack plane. This means that the dilatancy due to shear, observed
tween two different materials 共e.g., a matrix-inclusion interface兲, in most nonmetallic materials, is not taken into account 共at least
a plane of weakness 共e.g., a rock joint兲, or a simplified represen- not on the element level, although some dilatancy gets produced
tation of a narrow localization zone 共e.g., of a fracture process globally if the sliding interelement crack has a zig-zag path兲.
zone or a shear band兲. Some formulations assume a nonzero ini-
tial elastic compliance of the cohesive zone, which may be either Discretization Techniques for Cohesive Models
physically motivated by the reduced stiffness of the interface Discretization techniques used in conjunction with cohesive mod-
layer as compared to a perfect bond between two materials, or els can be divided into two broad categories, depending on
considered as a purely numerical artifact corresponding to a whether the displacement discontinuity can occur only between
penalty-type enforcement of displacement continuity in the elastic adjacent elements, or can run across the finite-element mesh
range. Other formulations enforce displacement continuity di- along an arbitrary trajectory.
rectly, and allow a nonzero separation only after a certain initia- In the first category, one can distinguish methods that assume
tion condition has been met. an a priori given trajectory of the discontinuity, methods that trace
Cohesive models for which the traction-separation law begins the trajectory by continuous remeshing, and methods that admit a
with a positive slope 共i.e., models with a nonzero elastic compli- potential discontinuity between any two adjacent finite elements.
ance, or models with softening preceded by hardening兲 must be The second category consists of finite-element formulations
used with care. If the stress increases above the initiation limit f 0 with embedded discontinuities, based on the enhanced assumed
共which is equal to zero for models with a nonzero elastic compli- strain method, and extended finite elements based on the
ance兲, it must also increase above f 0 at some material points partition-of-unity method.
sufficiently close to the crack face, yet a new cohesive crack In analogy to elements with embedded localization bands,
cannot initiate at that point. Because of this inconsistency, cracks which use discontinuous strain approximations, it is possible to
described by such cohesive models cannot be assumed to initiate construct elements with embedded localization lines 共in 2D兲 or
at an arbitrary location. Their path must be specified in advance or planes 共in 3D兲, across which the displacement field is discontinu-
restricted to a predefined finite number of segments. ous 共Dvorkin et al. 1990; Klisinski et al. 1991; Olofsson et al.
Needleman’s original model is based on an elastic potential 1994; Simo and Oliver 1994; Oliver 1996兲. For this numerical
without any internal variables, i.e., it has the character of a non- scheme, the trajectory of the cohesive crack or cohesive zone is

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1141

J. Eng. Mech., 2002, 128(11): 1119-1149

 independent of the finite-element mesh. This technique is much 2. on the tensile side of a reinforced concrete beam, where
more flexible than the standard scheme with discontinuities al- distributed cracking is stabilized by the tensile reinforce-
lowed only at element interfaces, and it eliminates the need for ment;
continuous remeshing. 3. in a system of cooling 共or drying兲 cracks propagating into a
A vast majority of embedded crack formulations developed in halfspace, especially if the temperature profile has a steep
the 1990s use a nonconforming approximation of the displace- front 共Bažant and Ohtsubo 1977; Bažant and Cedolin 1991,
ment jump. On one hand, this is convenient, because the approxi- Sec. 12.5兲;
mation of the displacement jump in one element can be com- 4. in dynamic problems when stress waves travel across the
pletely decoupled from the other elements and eliminated on the structure; and
local level. However, the elements then become sensitive to the 5. if heterogeneity of the material is explicitly taken into ac-
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

count as the mesh is refined.

orientation of the discontinuity line with respect to the nodes, and
Mesh refinement can cause parallel potential cohesive cracks to
in some unfavorable situations the response of the element ceases
become arbitrarily close. Thus, if they are under the same stress,
to be unique 共Jirásek 2000b兲.
their cumulative opening displacements per unit volume of mate-
A much more robust implementation can be achieved if the
rial 共if uniformly stressed兲 can become arbitrarily large in such
discontinuous enrichment is based on the partition-of-unity con-
situations. Such behavior was illustrated by Bažant’s example
cept 共Melenk and Babuška 1996; Duarte and Oden 1996兲, with a 共Bažant 1985a,b, 1986兲 of a concrete bar under uniaxial tension,
conforming approximation of the displacement jump and com- with strong enough axial steel reinforcement such that the overall
pletely independent strain components in the two parts of the tangential stiffness of the bar never becomes negative, despite
element separated by the crack 共Wells 2001兲. This formulation is softening in the cracks 共same as Fig. 14兲.
intimately related to the manifold method, developed by Shi Consequently, cohesive models with an assumed mesh-
共1991兲 for preexisting material discontinuities such as rock joints, independent softening law are applicable only when the strain-
and to the extended finite-element method 共XFEM兲, developed by softening damage is a priori known to localize. In a certain sense,
Belytschko and co-workers for stress-free cracks and material in- they are complementary to continuum-based models, for which a
terfaces 共Moës et al. 1999; Sukumar et al. 2000; Daux et al. mesh-independent stress–strain law with softening can cover only
2000兲 and adapted by Moës and Belytschko 共2002兲 for cohesive the cases when cracking remains perfectly distributed. In a gen-
cracks. eral case with regions of diffuse and localized cracking 共and per-
The traction-separation laws for embedded discontinuities are haps with gradual transition from diffuse damage to localized
usually postulated in a plasticity format 共Simo and Oliver 1994; fracture兲, both classes of models would need an adjustment taking
Oliver 1996; Olofsson et al. 1994; Armero and Garikipati 1995; into account the element size, shape, and orientation.
Larsson et al. 1996; Ohlsson and Olofsson 1997兲 or constructed For example, the model developed at LCPC Paris 共Rossi and
as extensions of the cohesive crack model to mixed-mode situa- Wu 1992兲 works with elastic elements separated by cohesive in-
tions 共Dvorkin et al. 1990; Klisinski et al. 1991; Lotfi and Shing terfaces, and the material parameters describing those interfaces
1995兲 using concepts similar to fixed crack versions of smeared are considered as random variables. To obtain mesh-independent
crack models. However, the plasticity-based models do not prop- results, at least in the global sense, it is necessary to adjust the
erly describe unloading of a brittle material. This problem be- statistical characteristics of the random strength distribution to the
comes especially severe at late stages of the degradation process size of the elastic elements. For small elements, the strength char-
when the crack is stress free and, according to the plasticity acterizes the properties on a small scale, and it has a large varia-
theory, a reversal of the opening rate immediately generates a tion. For large elements, the strength should characterize the over-
compressive traction, which is not physically realistic. The theo- all behavior on a larger scale, resulting from the combined effect
ries inspired by the cohesive crack model for concrete are close to of small-scale processes that cannot be captured explicitly on the
interface damage mechanics 共Simo and Oliver 1994; Oliver 1996; given resolution level. Experimental size effect investigations in-
Armero 1997兲, which provides a natural description of the dicate that both the mean value and the variation of strength de-
gradual loss of integrity. In this framework, a difficult issue is the crease as the size of the sample or specimen increases. The em-
proper treatment of stiffness recovery upon complete crack clo- pirically derived rule for the size dependence of the statistical
sure, with possible frictional sliding. Such effects were consis- characteristics of random strength distribution developed at LCPC
tently taken into account by Cangemi et al. 共1996兲 and by Chabo- seems to give globally objective results when used in simulations
che and coworkers 共Chaboche et al. 1997a,b兲 in the context of on different meshes.
Another limitation of the cohesive cracks is that they cannot
interface models for delamination and debonding of fiber–matrix
capture stress multiaxiality in the fracture process zone. The usual
composites. For cracks, a similar approach was proposed by
cohesive crack or cohesive zone models postulate a traction-
Jirásek 共1998a兲 and Jirásek and Zimmermann 共2001兲.
separation relation that takes into account only the out-of-plane
components of the stress tensor at the discontinuity surface. This
Limitations of Cohesive Models approach suits just fine all the notched fracture specimens in com-
The cohesive crack or cohesive zone models provide an objective mon use but is not representative of many applications. The effect
description of fully localized failure. In this case, the cohesive of the compressive normal stresses acting parallel to the crack
traction-separation law with softening does not need any adjust- plane 共called sometimes the ‘‘T stresses’’兲 is ignored. The fact that
ment for the element size, because mesh refinement does not this cannot be universally correct is clear from the observation
change the resolved crack pattern. However, the mesh indepen- that if one of the T stresses approaches the uniaxial compressive
dence is questionable if the cracking pattern is diffuse. This can strength f c of the material, the strength limit of the cohesive crack
happen, e.g., model should approach zero. Thus, for general situations, a mul-
1. in concrete with a sufficiently dense and strong reinforcing tiaxial cohesive crack model would have to be developed. Even
net; though this would, in principle, be possible, the multiaxial behav-

1142 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 ior in the fracture process zone can be captured more easily by Nonlocal Probabilistic Models of Failure
using a nonlocal model or crack band model with a good triaxial
constitutive law. It is impossible to do justice to the subject without pointing out
Finally, since the softening damage in the early stage of open- the importance of nonlocality in probabilistic modeling of failure.
ing of a cohesive crack occurs within a band of a certain width, The evolution of the classical probabilistic theory of strength,
the cohesive stress at location x should depend not merely on the begun in qualitative terms by Mariotte 共1686兲, culminated with
crack opening at x but on the average of crack opening taken over the studies on extreme value statistics and discovery of Weibull
a certain neighborhood of x. In other words, the cohesive law distribution by Fischer and Tippett 共1928兲, and the application of
itself may need to be considered as nonlocal 共Bažant 2001, this distribution to fracture by Weibull 共1939兲. This theory rests
2002b兲, which was taken into account in the gradient-enhanced on the hypothesis that the structure fails if a material element that
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

formulation of the cohesive crack model proposed by van Gils is infinitesimal compared to structure size D fails. The theory
共1997兲. works just fine for fatigue-embrittled metals and fine-grained ce-
ramics. It is noteworthy that since the size effect of this statistical
theory is a power law, self-similarity is implied, and no charac-
Regularization by Real or Artificial Viscosity
teristic structure size D 0 exists. The model possesses no charac-
In dynamics, nonlocal averaging can sometimes be avoided, as a teristic length. This is not realistic for quasibrittle materials.
convenient approximation, by considering material viscosity, rate This deficiency gets manifested in structures that do not fail
effect, or damping. Introducing artificial viscosity may serve as an immediately after the failure of one infinitesimal element. It re-
expedient substitute for nonlocality. quires modification for quasibrittle structures, consisting of mate-
Indeed, since the dimension of viscosity ␩ is kg/共m s兲, the rials such as concrete, composites, or ice, which are characterized
dimension of Young’s modulus E 共or strength兲 is kg/共m s2兲, and by a sizable fracture process zone 共FPZ兲, not negligible compared
the dimension of mass density ␳ is kg/m3, there exists a material to the structure size D. Two types of statistical behavior must be
length associated with viscosity, given by distinguished:
1. The structure fails only after a long fracture or a long dam-
␩ ␩ age band develops, which is typical of reinforced concrete, fiber
ᐉ v⫽ ⫽ (81) composites, and compression dominated problems. The path of
v ␳ 冑E␳
the fracture or band and the location of the FPZ are dictated
where v ⫽ 冑E/␳⫽longitudinal wave propagation velocity. Conse-
mainly by mechanics on the macroscale and are little sensitive to
material randomness when the structure is large. Within the FPZ
quently, any rate dependence in the constitutive law combined
the material behaves randomly, but if the size of the FPZ, which
with inertial effects introduces a length scale. This effect was
is of the same order as the intrinsic material length ᐉ, is negligible
exploited as a localization limiter regularizing the boundary value
compared to structure size D 共i.e., if DⰇᐉ), the statistical influ-
problem, for example, by Needleman 共1988兲. In a similar fashion,
ence disappears and the size effect is purely deterministic 共Bažant
a length scale is introduced by the delay effects in the damage
and Xi 1991兲. The classical Weibull theory cannot treat the large
evolution law 共Ladevèze 1992, 1995; Ladevèze et al. 2000兲.
size limit for which the fracture or band front, in relative coordi-
There is, however, an important difference from the nonlocal
nates, becomes perfectly sharp. The reason is that the Weibull
models—the viscosity-induced nonlocality gradually disappears
integral, see Eq. 共83兲 below, diverges for a singular stress distri-
with the passage of time. For load durations much larger than the
bution, which is approached as D/ᐉ→⬁.
relaxation 共or retardation兲 time associated with the type of viscos-
2. The structure fails at the initiation of fracture from the sur-
ity used, the modeling is not completely objective. Thus, the vis-
face, which is exemplified by flexural failure of an unreinforced
cosity or rate effect can be used as a substitute for a nonlocal
beam. If the structure is not large compared to ᐉ, the stress redis-
model only within a narrow range of time delays and rates, gen-
tribution in the cross section engendered by the FPZ is important
erally not spreading more than one order of magnitude. Therefore,
and dominates the failure load. The statistical effects are then
if an artificial viscosity is used, cautious insight is needed.
negligible, and the classical Weibull statistical theory does not
If a characteristic length, ᐉ v , is considered in dynamic prob-
apply. However, for DⰇᐉ, the FPZ relative to D becomes a point,
lems, then a characteristic time ␶ 0 is automatically implied as the
and the structure fails as soon as the first point fails. Since this
time of passage of a wave front over the distance ᐉ v :
point can have many random locations, the statistical effect on
ᐉv ␩ ␩ failure is important, and in this limit case the Weibull theory does
␶ 0⫽ ⫽ 2 ⫽ (82) apply. The larger the structure, the smaller is the strength that can
v v ␳ E
be encountered by this randomly located point.
In problems of impact of missiles on concrete walls, for ex- The nonlocal concept has been shown to provide a remedy to the
ample, the duration of the dynamic event is not too long com- abovementioned problems. In the simplest theory, which deals
pared to ␶ 0 . Therefore, nonlocal averaging is generally not only with the failure load and ignores the randomness of the
needed. Due to inertia effects, there is not enough time for spuri- deformations and stresses prior to failure, the probability of struc-
ous localization of strain-softening damage to develop, whether or tural failure, p f , is given by the Weibull integral whose integrand
not any viscosity or damping is taken into account. involves the 共local兲 maximum principal stress ␴ I (x). The nonlo-
If the viscosity is introduced in the constitutive law in such a cal remedy simply consists in replacing ␴ I (x) with the nonlocal
manner that there is no softening for fast enough loading 关e.g., maximum principal stress ˜␴ (x). Thus, the Weibull integral over
Bažant and Li’s 共1997兲 model for the rate dependence or cohesive structure volume 共in the case of tensile failures兲 takes the form
cracks in concrete兴, then the boundary value problem is regular 共Bažant and Xi 1991; Bažant and Novák 2000b兲:

再 冕冓 冔 冎
共well posed兲 for fast enough loading even in the absence of inertia
˜␴ I 共 x兲 m
dx
forces 共i.e., in quasistatics兲, simply because there is no strain soft- p f ⫽1⫺exp ⫺ (83)
ening in the structure. V ␴0 Vr

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1143

J. Eng. Mech., 2002, 128(11): 1119-1149

Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

Fig. 19. Logarithmic size effect plots of 共a兲 the numerical results obtained with the nonlocal Weibull-type probabilistic theory, and 共b兲 previously
published test data on the modulus of rupture of concrete, f r 共i.e., flexural strength of unreinforced beams兲, and their fit by an asymptotic formula
关in 共b兲, the optimum fit is obtained first for each data set separately and then the data are plotted in relative coordinates兴; after Bažant and Novák
共2000a兲.

where m and ␴ 0 are material constants; m is the Weibull modulus methods do not meet this requirement, and thus their use for
(m⬇24 for concrete兲, and ␴ 0 is the Weibull scale parameter, with predicting failure loads of extremely small probability is doubtful.
the dimension of stress. The nonlocal stress cannot be defined Formulation of a realistic general statistical nonlocal theory is a
simply as the average stress. Rather, it must be defined in terms of challenge for the future.
the nonlocal inelastic stress or nonlocal inelastic strain. Choosing
the latter, one has
Concluding Thoughts and Future Path
␴
˜ ⫽De 共 ⑀⫺⑀
¯i 兲 (84)
While two decades ago strain-softening damage models were re-
where ⑀i is the inelastic strain and ¯⑀i is its nonlocal average. For
garded as controversial, the nonlocal concept has by now ren-
details, consult Bažant and Novák 共2000b,c兲, who showed that the
dered them respectable and their use realistic. Nonlocality is now
theory gives good agreement with extensive test data regarding
generally accepted as the proper approach for regularizing the
the combined probabilistic-energetic size effect on the modulus of
boundary problems of continuum damage mechanics, for captur-
rupture of unreinforced concrete beams 共Bažant and Novák
ing the size effect, and for avoiding spurious localization, giving
2000a兲. The logarithmic size effect plots of the numerical results
rise to pathological mesh sensitivity.
obtained with the nonlocal Weibull-type probabilistic theory are
The last two decades of research gave birth to a wide variety
shown in Fig. 19共a兲. Note that a deterministic nonlocal theory
of nonlocal models, with differences that are not quite justified by
gives almost the same results for the left half of the plot but
diversity in the types of materials and practical application. One
terminates with a horizontal asymptote. The inclined large-size
may now expect a period of crystalization in which many artifi-
asymptote corresponds to the classical Weibull statistical size ef-
cially complex or oversimplified models will fade, being recog-
fect without nonlocality. The combination of nonlocality and sta-
nized as superfluous, and only a few will gain a permanent ped-
tistics is needed for the transition, through the intermediate sizes.
estal in the pantheon of knowledge.
Evidently, amalgamation of the statistical and nonlocal theories is
Speculations though could be made in this regard, they are
crucial for correct extrapolation from small-scale laboratory tests
better left to future scrutiny. The strength of each model’s pedes-
to large structures 共such as the vertical bending fracture of an arch
tal will be judged by the physics of the microstructure, and the
dam兲.
permanence of that pedestal will be decided by passage of each
A much more difficult proposition is the development of a
model through the sieve of practical applications. Doubtless much
nonlocal probabilistic theory that would give not only the failure
further research lies ahead, and polemics will enliven the path
probabilities, independent of the loss of positive definiteness of
into the future.
the structural stiffness matrix, but would also describe the prob-
ability distributions of deflections and stresses prior to failure.
The theory must yield the probability distribution of the first ei- Acknowledgments
genvalue ␭ 1 of the structural stiffness matrix as the failure is
approached. The main difficulty is that what matters is the distri- Partial financial support of the first writer has been obtained under
bution tail, with probabilities of the order of 10⫺7 . For D→⬁, the Grant No. N00014-91-J-1109 from the Office of Naval Research
tail of the distribution of ␭ 1 as ␭ 1 →0 must have the form of 共program director Y. D. S. Rajapakse兲 to Northwestern University,
Weibull distribution when the relative structure size D/ᐉ is ap- and financial support of the second writer has been provided by
proaching infinity, because in that limit the classical Weibull the Swiss Commission for Technology and Innovation under
theory must be recovered. The existing stochastic finite-element Grant Nos. CTI 4424.1 and CTI 5501.1.

1144 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 References Bažant, Z. P., and Christensen, M. 共1972a兲. ‘‘Analogy between micropo-
lar continuum and grid frameworks under initial stress.’’ Int. J. Solids
Struct., 8, 327–346.
Aero, E. L., and Kuvshinskii, E. V. 共1960兲. ‘‘Fundamental equations of
Bažant, Z. P., and Christensen, M. 共1972b兲. ‘‘Long-wave extensional
the theory of elastic materials with rotationally interacting particles.’’
buckling of large regular frames.’’ J. Struct. Div., ASCE, 98共10兲,
Fiz. Tverd. Tela (S.-Peterburg), 2, 1399–1409.
2269–2289.
Armero, F. 共1997兲. ‘‘Localized anisotropic damage of brittle materials.’’
Bažant, Z. P., and Jirásek, M. 共1994兲. ‘‘Nonlocal model based on crack
Computational plasticity: Fundamentals and applications, D. R. J.
Owen, E. Oñate, and E. Hinton, eds., International Center for Numeri- interactions: A localization study.’’ J. Eng. Mater. Technol., 116, 256 –
cal Methods in Engineering, Barcelona, Spain, 635– 640. 259.
Armero, F., and Garikipati, K. 共1995兲. ‘‘Recent advances in the analysis Bažant, Z. P., and Li, Y.-N. 共1995兲. ‘‘Stability of cohesive crack model.’’
and numerical simulation of strain localization in inelastic solids.’’ J. Appl. Mech., 62, 959–969.
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

Computational plasticity: Fundamentals and applications, D. R. J. Bažant, Z. P., and Li, Y.-N. 共1997兲. ‘‘Cohesive crack model with rate-
Owen and E. Oñate, eds., International Center for Numerical Methods dependent crack opening and viscoelasticity: I. Mathematical model
in Engineering, Barcelona, Spain, 547–561. and scaling.’’ Int. J. Fract. 86, 247–265.
Barenblatt, G. I. 共1962兲. ‘‘The mathematical theory of equilibrium of Bažant, Z. P., and Lin, F.-B. 共1988a兲. ‘‘Nonlocal smeared cracking model
cracks in brittle fracture.’’ Adv. Appl. Mech., 7, 55–129. for concrete fracture.’’ J. Struct. Eng., 114共11兲, 2493–2510.
Bažant, Z. P. 共1971兲. ‘‘Micropolar medium as a model for buckling of Bažant, Z. P., and Lin, F.-B. 共1988b兲. ‘‘Nonlocal yield-limit degradation.’’
grid frameworks. Developments in mechanics.’’ Proc., 12th Midwest- Int. J. Numer. Methods Eng., 26, 1805–1823.
ern Mechanics Conference, Univ. of Notre Dame, 587–593. Bažant, Z. P., and Novák, D. 共2000a兲. ‘‘Energetic-statistical size effect in
Bažant, Z. P. 共1976兲. ‘‘Instability, ductility, and size effect in strain- quasibrittle failure at crack initiation.’’ ACI Mater. J., 97, 381–392.
softening concrete.’’ J. Eng. Mech. Div., 102共2兲, 331–344. Bažant, Z. P., and Novák, D. 共2000b兲. ‘‘Probabilistic nonlocal theory for
Bažant, Z. P. 共1982兲. ‘‘Crack band model for fracture of geomaterials.’’ quasibrittle fracture initiation and size effect: I. Theory.’’ J. Eng.
Proc., 4th Int. Conf. on Numerical Methods in Geomechanics, Z. Mech., 126共2兲, 166 –174.
Eisenstein, ed., Univ. of Alberta, Edmonton, 3, 1137–1152. Bažant, Z. P., and Novák, D. 共2000c兲. ‘‘Probabilistic nonlocal theory for
Bažant, Z. P. 共1984a兲. ‘‘Imbricate continuum and its variational deriva- quasibrittle fracture initiation and size effect: II. Application.’’ J. Eng.
tion.’’ J. Eng. Mech., 110共12兲, 1693–1712. Mech., 126共2兲, 175–185.
Bažant, Z. P. 共1984b兲. ‘‘Size effect in blunt fracture: Concrete, rock, and Bažant, Z. P., and Oh, B.-H. 共1983兲. ‘‘Crack band theory for fracture of
metal.’’ J. Eng. Mech., 110共4兲, 518 –535. concrete.’’ Mater. Struct., 16, 155–177.
Bažant, Z. P. 共1985a兲. ‘‘Fracture in concrete and reinforced concrete.’’ Bažant, Z. P., and Ohtsubo, H. 共1977兲. ‘‘Stability conditions for propaga-
Mechanics of geomaterials: Rocks, concretes, and soils, Z. P. Bažant, tion of a system of cracks in a brittle solid.’’ Mech. Res. Commun., 4,
ed., Wiley, London, 259–303. 353–366.
Bažant, Z. P. 共1985b兲. ‘‘Mechanics of fracture and progressive cracking Bažant, Z. P., and Ožbolt, J. 共1990兲. ‘‘Nonlocal microplane model for
concrete structures.’’ Fracture mechanics of concrete: Structural ap- fracture, damage, and size effect in structures.’’ J. Eng. Mech.,
plication and numerical calculation, G. C. Sih and A. DiTommaso, 116共11兲, 2485–2505.
eds., Martinus Nijhoff, Dordrecht, The Netherlands, 1, 1–94. Bažant, Z. P., and Pfeiffer, P. A. 共1987兲. ‘‘Determination of fracture en-
Bažant, Z. P. 共1986兲. ‘‘Mechanics of distributed cracking.’’ Appl. Mech. ergy from size effect and brittleness number.’’ ACI Mater. J., 84, 463–
Rev., 39, 675–705. 480.
Bažant, Z. P. 共1987兲. ‘‘Why continuum damage is nonlocal: Justification Bažant, Z. P., and Pijaudier-Cabot, G. 共1988兲. ‘‘Nonlocal continuum dam-
by quasiperiodic microcrack array.’’ Mech. Res. Commun., 14, 407– age, localization instability and convergence.’’ J. Appl. Mech., 55,
419. 287–293.
Bažant, Z. P. 共1991兲. ‘‘Why continuum damage is nonlocal: Microme- Bažant, Z. P., and Pijaudier-Cabot, G. 共1989兲. ‘‘Measurement of charac-
chanics arguments.’’ J. Eng. Mech., 117共5兲, 1070–1087. teristic length of nonlocal continuum.’’ J. Eng. Mech., 115共4兲, 755–
Bažant, Z. P. 共1994兲. ‘‘Nonlocal damage theory based on micromechanics 767.
of crack interactions.’’ J. Eng. Mech., 120共3兲, 593– 617. Bažant, Z. P., and Planas, J. 共1998兲. Fracture and size effect in concrete
Bažant, Z. P. 共2001兲. ‘‘Concrete fracture models: Testing and practice.’’ and other quasibrittle materials, CRC, Boca Raton, Fla.
Eng. Fract. Mech., 69, 165–206. Bažant, Z. P., Tabbara, M. R., Kazemi, M. T., and Pijaudier-Cabot, G.
Bažant, Z. P. 共2002a兲. ‘‘Reminiscences on four decades of struggle and 共1990兲. ‘‘Random particle model for fracture of aggregate or fiber
progress in softening damage and size effect.’’ Concr. J. (Japan composites.’’ J. Eng. Mech., 116共8兲, 1686 –1705.
Concr. Inst.), 40, 16 –28. Bažant, Z. P., and Xi, Y. 共1991兲. ‘‘Statistical size effect in quasibrittle
Bažant, Z. P. 共2002b兲. Scaling of structural strength, Hermes-Penton, structures: II. Nonlocal theory.’’ J. Eng. Mech., 117共11兲, 2623–2640.
London. Bažant, Z. P., Xiang, Y., Adley, M., Prat, P. C., and Akers, S. A. 共1996兲.
Bažant, Z. P., and Belytschko, T. B. 共1985兲. ‘‘Wave propagation in a ‘‘Microplane model for concrete: II: Data delocalization and verifica-
strain-softening bar: Exact solution.’’ J. Eng. Mech., 111共3兲, 381–389. tion.’’ J. Eng. Mech., 122共3兲, 255–262.
Bažant, Z. P., Belytschko, T. B., and Chang, T.-P. 共1984兲. ‘‘Continuum Belytschko, T., Fish, J., and Engelmann, B. E. 共1988兲. ‘‘A finite-element
model for strain softening.’’ J. Eng. Mech., 110共12兲, 1666 –1692. with embedded localization zones.’’ Comput. Methods Appl. Mech.
Bažant, Z. P., Caner, F. C., Carol, I., Adley, M. D., and Akers, S. 共2000兲. Eng., 70, 59– 89.
‘‘Microplane model M4 for concrete: I. Formulation with work- Benvenuti, E., Borino, G., and Tralli, A. 共2000兲. ‘‘A thermodynamically
conjugate deviatoric stress.’’ J. Eng. Mech., 126共9兲, 944 –953. consistent nonlocal formulation for elastodamaging materials: Theory
Bažant, Z. P., and Cedolin, L. 共1979兲. ‘‘Blunt crack band propagation in and computations.’’ Proc., ECCOMAS 2000, Barcelona, Spain, CD-
finite element analysis.’’ J. Eng. Mech., 105共2兲, 297–315. ROM.
Bažant, Z. P., and Cedolin, L. 共1991兲. Stability of structures, Oxford Besdo, D. 共1974兲. ‘‘Ein Beitrag zur nichtlinearen Theorie des Cosserat
University Press, New York, Chap. 10. Kontinuums.’’ Acta Mech., 20, 105–131.
Bažant, Z. P., Červenka, J., and Wierer, M. 共2001兲. ‘‘Equivalent localiza- Borino, G., and Failla, B. 共2000兲. ‘‘Thermodynamically consistent plas-
tion element for crack band model and as alternative to elements with ticity models with local and nonlocal internal variables.’’ Proc., Eu-
embedded discontinuities.’’ Fracture mechanics of concrete struc- ropean Congr. on Computational Methods in Applied Sciences and
tures, R. de Borst et al., ed., Balkema, Lisse, The Netherlands, 765– Engineering, ECCOMAS, Barcelona, Spain, CD-ROM.
772. Borino, G., Failla, B., and Parrinello, F. 共2002兲. ‘‘A symmetric formula-
Bažant, Z. P., and Chang, T.-P. 共1984兲. ‘‘Instability of nonlocal continuum tion for nonlocal damage models.’’ H. A. Mang, F. G. Rammerstorfer,
and strain averaging.’’ J. Eng. Mech., 110共10兲, 1441–1450. and J. Eberhardsteiner, eds., Proc., 5th World Congress on Computa-

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1145

J. Eng. Mech., 2002, 128(11): 1119-1149

 tional Mechanics (WCCM V), Vienna Univ. of Technology, Vienna, ement size for elastic composites.’’ J. Mech. Phys. Solids, 44, 497–
Austria, ISBN 3-9501554-0-6, 具http://wccm.tuwien.ac.at典. 524.
Borino, G., Fuschi, P., and Polizzotto, C. 共1999兲. ‘‘A thermodynamic Duarte, C. A., and Oden, J. T. 共1996兲. ‘‘H–p clouds—An h–p meshless
approach to nonlocal plasticity and related variational approaches.’’ J. method.’’ Numer. Methods Partial Diff. Eq., 12, 673–705.
Appl. Mech., 66, 952–963. Dugdale, D. S. 共1960兲. ‘‘Yielding of steel sheets containing slits.’’ J.
Borino, G., and Polizzotto, C. 共1999兲. ‘‘Comments on ‘Nonlocal bar re- Mech. Phys. Solids, 8, 100–108.
visited by Christer Nilsson’.’’ Int. J. Solids Struct., 36, 3085–3091. Duhem, P. 共1893兲. ‘‘Le potentiel thermodynamique et la pression hy-
Boudon-Cussac, D., Hild, F., and Pijaudier-Cabot, G. 共1999兲. ‘‘Tensile drostatique.’’ Ann. Sci. Ec. Normale Super., 10, 183–230.
damage in concrete: Analysis of experimental technique.’’ J. Eng. Dvorkin, E. N., Cuitiño, A. M., and Gioia, G. 共1990兲. ‘‘Finite elements
Mech., 125共8兲, 906 –913. with displacement interpolated embedded localization lines insensi-
Camacho, G. T., and Ortiz, M. 共1996兲. ‘‘Computational modeling of im- tive to mesh size and distortions.’’ Comput. Methods Appl. Mech.
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

pact damage in brittle materials.’’ Int. J. Solids Struct., 33, 2899– Eng., 90, 829– 844.
2938. Edelen, D. G. B. 共1969兲. ‘‘Protoelastic bodies with large deformation.’’
Cangemi, L., Cocu, M., and Raous, M. 共1996兲. ‘‘Adhesion and friction Arch. Ration. Mech. Anal., 34, 283–300.
model for the fiber/matrix interface of a composite.’’ Proc., 1996 En- Edelen, D. G. B., Green, A. E., and Laws, N. 共1971兲. ‘‘Nonlocal con-
gineering Systems Design and Analysis Conf., 157–163. tinuum mechanics.’’ Arch. Ration. Mech. Anal., 43, 36 – 44.
Carmeliet, J. 共1999兲. ‘‘Optimal estimation of gradient damage parameters Edelen, D. G. B., and Laws, N. 共1971兲. ‘‘On the thermodynamics of
from localization phenomena in quasibrittle materials.’’ Mech. systems with nonlocality.’’ Arch. Ration. Mech. Anal., 43, 24 –35.
Cohesive-Frict. Mater., 4, 1–16. Engelen, R. A. B., Geers, M. G. D., and Baaijens, F. P. T. 共2002兲. ‘‘Non-
Carol, I., and Bažant, Z. P. 共1997兲. ‘‘Damage and plasticity in microplane local implicit gradient-enhanced elastoplasticity for the modeling of
theory.’’ Int. J. Solids Struct., 34, 3807–3835. softening behavior.’’ Int. J. Plast., in press.
Carol, I., and Prat, P. C. 共1990兲. ‘‘A statically constrained microplane Ericksen, J. L. 共1960兲. ‘‘Anisotropic fluids.’’ Arch. Ration. Mech. Anal., 4,
model for the smeared analysis of concrete cracking.’’ in Computer 231–237.
aided analysis and design of concrete structures, N. Bićanić and H. Eringen, A. C. 共1964兲. ‘‘Simple microfluids.’’ Int. J. Eng. Sci., 2, 205–
Mang, eds., Pineridge, Swansea, U.K., 919–930. 217.
Cedolin, L., and Bažant, Z. P. 共1980兲. ‘‘Effect of finite-element choice in Eringen, A. C. 共1966a兲. ‘‘Linear theory of micropolar elasticity.’’ J. Math.
blunt crack band analysis.’’ Comput. Methods Appl. Mech. Eng., 24, Mech., 15, 909–924.
305–316. Eringen, A. C. 共1966b兲. ‘‘Mechanics of micromorphic materials.’’ Proc.,
Červenka, J. 共1994兲. ‘‘Discrete crack modeling in concrete structures.’’ 11th Int. Congr. of Applied Mechanics, Springer, Berlin, 131–138.
PhD thesis, Univ. of Colorado, Boulder, Colo. Eringen, A. C. 共1966c兲. ‘‘A unified theory of thermomechanical materi-
Červenka, J., Bažant, Z. P., and Wierer, M. 共2002兲. ‘‘Equivalent localiza- als.’’ Int. J. Eng. Sci., 4, 179–202.
tion element for crack band approach to mesh-size sensitivity in mi- Eringen, A. C. 共1972兲. ‘‘Linear theory of nonlocal elasticity and disper-
croplane model.’’ Int. J. Numer. Methods Eng., in press. sion of plane waves.’’ Int. J. Eng. Sci., 10, 425– 435.
Chaboche, J. L., Girard, R., and Levasseur, P. 共1997a兲. ‘‘On the interface Eringen, A. C. 共1981兲. ‘‘On nonlocal plasticity.’’ Int. J. Eng. Sci., 19,
debonding models.’’ Int. J. Damage Mech., 6, 220–257. 1461–1474.
Chaboche, J. L., Girard, R., and Schaff, A. 共1997b兲. ‘‘Numerical analysis Eringen, A. C. 共1983兲. ‘‘Theories of nonlocal plasticity.’’ Int. J. Eng. Sci.,
of composite systems by using interphase/interface models.’’ Compu- 21, 741–751.
tational Mech., Berlin, 20, 3–11. Eringen, A. C., and Ari, N. 共1983兲. ‘‘Nonlocal stress field at a Griffith
Chandrasekhar, S. 共1950兲. Radiation transfer, Oxford University Press, crack.’’ Cryst. Lattice Defects Amorphous Mater., 10, 33–38.
London. Eringen, A. C., and Edelen, D. G. B. 共1972兲. ‘‘On nonlocal elasticity.’’
Comi, C. 共2001兲. ‘‘A nonlocal model with tension and compression dam- Int. J. Eng. Sci., 10, 233–248.
age mechanisms.’’ Eur. J. Mech. A/Solids, 20, 1–22. Eringen, A. C., and Kim, B. S. 共1974兲. ‘‘Stress concentration at the tip of
Comi, C., and Perego, U. 共2001兲. ‘‘Numerical aspects of nonlocal damage a crack.’’ Mech. Res. Commun., 1, 233–237.
analyses.’’ Rev. Europ. Elements Finis, 10, 227–242. Eringen, A. C., Speziale, C. G., and Kim, B. S. 共1977兲. ‘‘Crack-tip prob-
Cope, R. J., Rao, P. V., Clark, L. A., and Norris, P. 共1980兲. ‘‘Modeling of lem in nonlocal elasticity.’’ J. Mech. Phys. Solids, 25, 339–355.
reinforced concrete behavior for finite-element analysis of bridge Eringen, A. C., and Suhubi, E. S. 共1964兲. ‘‘Nonlinear theory of simple
slabs.’’ Numerical methods for nonlinear problems, microelastic solids.’’ Int. J. Eng. Sci., 2, 189–203, 389– 404.
Pineridge, Swansea, 1, 457– 470. Etse, G., and Willam, K. J. 共1994兲. ‘‘Fracture-energy formulation for
Cordebois, J. P., and Sidoroff, F. 共1979兲. ‘‘Anisotropie élastique induite inelastic behavior of plain concrete.’’ J. Eng. Mech., 120共9兲, 1983–
par endommagement.’’ Comportement mécanique des solides aniso- 2011.
tropes, No. 295, Colloques internationaux du CNRS, Editions du Fabrikant, V. I. 共1990兲. ‘‘Complete solutions to some mixed boundary
CNRS, Grenoble, France, 761–774. value problems in elasticity.’’ Adv. Appl. Mech., 27, 153–223.
Cosserat, E., and Cosserat, F. 共1909兲. Théorie des corps déformables, Ferrara, L. 共1998兲. ‘‘A contribution to the modeling of mixed-mode frac-
Herrman, Paris. ture and shear transfer in plain and reinforced concrete.’’ PhD thesis,
Daux, C., Moës, N., Dolbow, J., Sukumar, N., and Belytschko, T. 共2000兲. Politecnico di Milano, Milano, Italy.
‘‘Arbitrary branched and intersecting cracks with the extended finite- Fischer, R. A., and Tippett, L. H. C. 共1928兲. ‘‘Limiting forms of the
element method.’’ Int. J. Numer. Methods Eng., 48, 1741–1760. frequency distribution of the largest and smallest member of a
de Borst, R. 共1986兲. ‘‘Nonlinear analysis of frictional materials.’’ PhD sample.’’ Proc. Cambridge Philos. Soc., 24, 180–190.
thesis, Delft Univ. of Technology, Delft, The Netherlands. Fish, J., and Yu, Q. 共2001兲. ‘‘Multiscale damage modeling for composite
Denarié, E., Saouma, V. E., Iocco, A., and Varelas, D. 共2001兲. ‘‘Concrete materials: Theory and computational framework.’’ Int. J. Numer.
fracture process zone characterization with fiber optics.’’ J. Eng. Methods Eng., 52, 161–191.
Mech., 127共5兲, 494 –502. Fleck, N. A., and Hutchinson, J. W. 共1993兲. ‘‘A phenomenological theory
di Prisco, M., and Mazars, J. 共1996兲. ‘‘Crush–crack: A nonlocal damage for strain gradient effects in plasticity.’’ J. Mech. Phys. Solids, 41,
model for concrete.’’ J. Mech. Cohesive Frict. Mater., 1, 321–347. 1825–1857.
Dı́ez, P., Egozcue, J. J., and Huerta, A. 共1998兲. ‘‘A posteriori error esti- Fleck, N. A., and Hutchinson, J. W. 共2001兲. ‘‘A reformulation of strain
mation for standard finite-element analysis.’’ Comput. Methods Appl. gradient plasticity.’’ J. Mech. Phys. Solids, 49, 2245–2271.
Mech. Eng., 163, 141–157. Fleck, N. A., Muller, G. M., Ashby, M. F., and Hutchinson, J. W. 共1994兲.
Drugan, W. J., and Willis, J. R. 共1996兲. ‘‘A micromechanics-based non- ‘‘Strain gradient plasticity: Theory and experiment.’’ Acta Metall.
local constitutive equation and estimates of representative volume el- Mater., 42, 475– 487.

1146 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 Gao, H., and Huang, Y. 共2001兲. ‘‘Taylor-based nonlocal theory of plastic- Jirásek, M., and Bažant, Z. P. 共1995兲. ‘‘Macroscopic fracture characteris-
ity.’’ Int. J. Solids Struct., 38, 2615–2637. tics of random particle systems.’’ Int. J. Fract., 69, 201–228.
Gao, H., Huang, Y., Nix, W. D., and Hutchinson, J. W. 共1999兲. Jirásek, M., and Bažant, Z. P. 共2001兲. Inelastic analysis of structures,
‘‘Mechanism-based strain gradient plasticity. I. Theory.’’ J. Mech. Wiley, Chichester, U.K.
Phys. Solids, 47, 1239–1263. Jirásek, M., and Patzák, B. 共2002兲. ‘‘Consistent tangent stiffness for non-
Geers, M. G. D., de Borst, R., and Brekelmans, W. A. M. 共1996a兲. ‘‘Com- local damage models.’’ Comput. Struct., in press.
puting strain fields from discrete displacement fields in 2D solids.’’ Jirásek, M., and Rolshoven, S. 共2002兲. ‘‘Comparison of integral-type non-
Int. J. Solids Struct., 33, 4293– 4307. local plasticity models for strain-softening materials.’’ Int. J. Eng.
Geers, M. G. D., Engelen, R. A. B., and Ubachs, R. J. M. 共2001兲. ‘‘On the Sci., in press.
numerical modeling of ductile damage with an implicit gradient- Jirásek, M., and Zimmermann, T. 共1997兲. ‘‘Rotating crack model with
enhanced formulation.’’ Rev. Euro. Eléments finis, 10, 173–191. transition to scalar damage: I. Local formulation, II. Nonlocal formu-
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

Geers, M. G. D., Peijs, T., Brekelmans, W. A. M., and de Borst, R. lation and adaptivity.’’ LSC Internal Rep. No. 97/01, Swiss Federal
共1996b兲. ‘‘Experimental monitoring of strain localization and failure Institute of Technology, Lausanne, Switzerland.
behavior of composite materials.’’ Compos. Sci. Technol., 56, 1283– Jirásek, M., and Zimmermann, T. 共1998a兲. ‘‘Analysis of rotating crack
1290. model.’’ J. Eng. Mech., 124共8兲, 842– 851.
Green, A. E. 共1965兲. ‘‘Micromaterials and multipolar continuum mechan- Jirásek, M., and Zimmermann, T. 共1998b兲. ‘‘Rotating crack model with
ics.’’ Int. J. Eng. Sci., 3, 533–537. transition to scalar damage.’’ J. Eng. Mech., 124共3兲, 277–284.
Green, A. E., and Naghdi, P. M. 共1965兲. ‘‘A general theory of an elastic– Jirásek, M., and Zimmermann, T. 共2001兲. ‘‘Embedded crack model: I.
plastic continuum.’’ Arch. Ration. Mech. Anal., 18, 251–281. Basic formulation.’’ Int. J. Numer. Methods Eng., 50, 1269–1290.
Green, A. E., and Rivlin, R. S. 共1964a兲. ‘‘Multipolar continuum mechan- Kachanov, M. 共1985兲. ‘‘A simple technique of stress analysis in elastic
ics.’’ Arch. Ration. Mech. Anal., 17, 113–147. solids with many cracks.’’ Int. J. Fract., 28, R11–R19.
Green, A. E., and Rivlin, R. S. 共1964b兲. ‘‘Simple force and stress multi- Kachanov, M. 共1987兲. ‘‘Elastic solids with many cracks: A simple method
poles.’’ Arch. Ration. Mech. Anal., 16, 325–353. of analysis.’’ Int. J. Solids Struct., 23, 23– 43.
Grioli, G. 共1960兲. ‘‘Elasticità asimmetrica.’’ Annali di matematica pura Kang, H. D. 共1997兲. ‘‘Triaxial constitutive model for plain and reinforced
ed applicata, Ser. IV, 50, 389– 417. concrete behavior.’’ PhD thesis, Univ. of Colorado, Boulder, Colo.
Günther, W. 共1958兲. ‘‘Zur Statik und Kinematik des Cosseratschen Kon- Kennedy, T. C., and Nahan, M. F. 共1996兲. ‘‘A simple nonlocal damage
tinuum.’’ Abh. Braunschweigischen Wissenschaftlichen Gesellschaft, model for predicting failure of notched laminates.’’ Compos. Struct.,
10, 195–213. 35, 229–236.
Gupta, A. K., and Akbar, H. 共1984兲. ‘‘Cracking in reinforced concrete Kennedy, T. C., and Nahan, M. F. 共1997兲. ‘‘A simple nonlocal damage
analysis.’’ J. Struct. Eng., 110共8兲, 1735–1746. model for predicting failure in a composite shell containing a crack.’’
Hadamard, J. 共1903兲. Leçons sur la propagation des ondes, Hermann, Compos. Struct., 39, 85–91.
Paris. Klisinski, M., Runesson, K., and Sture, S. 共1991兲. ‘‘Finite element with
Halphen, B., and Nguyen, Q. S. 共1975兲. ‘‘Sur les matériaux standards inner softening band.’’ J. Eng. Mech., 117共3兲, 575–587.
généralisés.’’ J. Mec., 14, 39– 63. Koiter, W. T. 共1964兲. ‘‘Couple stresses in the theory of elasticity.’’ Proc.
Hillerborg, A., Modéer, M., and Peterson, P. E. 共1976兲. ‘‘Analysis of K. Ned. Akad. Wet., Ser. B: Phys. Sci., 67, 17– 44.
crack propagation and crack growth in concrete by means of fracture Kröner, E. 共1966兲. ‘‘Continuum mechanics and range of atomic cohesion
mechanics and finite elements.’’ Cem. Concr. Res., 6, 773–782. forces.’’ Proc., 1st Int. Conf. on Fracture, T. Yokobori, T. Kawasaki,
Hodgkin, A. L. 共1964兲. The conduction of nervous impulse, Thomas, and J. Swedlow, eds., Japanese Society for Strength and Fracture of
Springfield, Ill. Materials, Sendai, Japan, 27.
Hu, X. Z., and Wittmann, F. H. 共2000兲. ‘‘Size effect on toughness induced Kröner, E. 共1967兲. ‘‘Elasticity theory of materials with long range cohe-
by cracks close to free surface.’’ Eng. Fract. Mech., 65, 209–211. sive forces.’’ Int. J. Solids Struct., 3, 731–742.
Huang, Y., Gao, H., Nix, W. D., and Hutchinson, J. W. 共2000兲. Kröner, E., and Datta, B. K. 共1966兲. ‘‘Nichtlokale Elastostatik: Ableitung
‘‘Mechanism-based strain gradient plasticity. II. Analysis.’’ J. Mech. aus der Gittertheorie.’’ Z. Phys., 196, 203–211.
Phys. Solids, 48, 99–128. Krumhansl, J. A. 共1965兲. ‘‘Generalized continuum field representation for
Huerta, A., Dı́ez, P., Rodrı́guez-Ferran, A., and Pijaudier-Cabot, G. lattice vibrations.’’ Lattice dynamics, R. F. Wallis, ed., Pergamon,
共1998兲. ‘‘Error estimation and finite-element analysis of softening sol- London, 627– 634.
ids.’’ Advances in adaptive computational methods in mechanics, P. Kunin, I. A. 共1966a兲. ‘‘Model of elastic medium simple structure with
Ladevèze and J. T. Oden, eds., Elsevier, Oxford, U.K., 333–348. spatial dispersion.’’ Prikl. Mat. Mekh., 30, 942 共in Russian兲.
Huerta, A., and Pijaudier-Cabot, G. 共1994兲. ‘‘Discretization influence on Kunin, I. A. 共1966b兲. ‘‘Theory of elasticity with spatial dispersion.’’ Prikl.
regularization by two localisation limiters.’’ J. Eng. Mech., 120共6兲, Mat. Mekh., 30, 866 共in Russian兲.
1198 –1218. Kunin, I. A. 共1968兲. ‘‘The theory of elastic media with microstructure and
Irwin, G. R. 共1958兲. ‘‘Fracture.’’ in Handbuch der Physik, S. Flügge, ed., the theory of dislocations.’’ Mechanics of generalized continua, E.
Springer, Berlin, VI, 551–590. Kröner, ed., Springer, Heidelberg, 321–329.
Jirásek, M. 共1998a兲. ‘‘Finite elements with embedded cracks.’’ LSC In- Labuz, J. F., Cattaneo, S., and Chen, L.-H. 共2001兲. ‘‘Acoustic emission at
ternal Rep. No. 98/01, Swiss Federal Institute of Technology, Lau- failure in quasibrittle materials.’’ Constr. Build. Mater., 15, 225–233.
sanne, Switzerland. Ladevèze, P. 共1992兲. ‘‘A damage computational method for composite
Jirásek, M. 共1998b兲. ‘‘Nonlocal models for damage and fracture: Com- structures.’’ Comput. Struct., 44, 79– 87.
parison of approaches.’’ Int. J. Solids Struct., 35, 4133– 4145. Ladevèze, P. 共1995兲. ‘‘A damage computational method for composites:
Jirásek, M. 共1999兲. ‘‘Comments on microplane theory.’’ Mechanics of Basic aspects and micromechanical relations.’’ Computational Mech.,
quasibrittle materials and structures, G. Pijaudier-Cabot, Z. Bittnar, Berlin, 17, 142–150.
and B. Gérard, eds., Hermès Science, Paris, 55–77. Ladevèze, P., Allix, O., Deü, J.-F., and Lévêque, D. 共2000兲. ‘‘A meso-
Jirásek, M. 共2000a兲. ‘‘Comparative study on finite elements with embed- model for localization and damage computation in laminates.’’ Com-
ded cracks.’’ Comput. Methods Appl. Mech. Eng., 188, 307–330. put. Methods Appl. Mech. Eng., 183, 105–122.
Jirásek, M. 共2000b兲. ‘‘Conditions of uniqueness for finite elements with Lakes, R. S. 共1986兲. ‘‘Experimental microelasticity of two porous solids.’’
embedded cracks.’’ Proc., 6th Int. Conf. on Computational Plasticity, Int. J. Solids Struct., 22, 55– 63.
Barcelona. CD-ROM. Landis, E. N. 共1999兲. ‘‘Micro–macro-fracture relationships and acoustic
Jirásek, M., and Bažant, Z. P. 共1994兲. ‘‘Localization analysis of nonlocal emissions in concrete.’’ Constr. Build. Mater., 13, 65–72.
model based on crack interactions.’’ J. Eng. Mech., 120共7兲, 1521– Larsson, R., Runesson, K., and Sture, S. 共1996兲. ‘‘Embedded localization
1542. band in undrained soil based on regularized strong discontinuity—

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1147

J. Eng. Mech., 2002, 128(11): 1119-1149

 Theory and FE analysis.’’ Int. J. Solids Struct., 33, 3081–3101. Oliver, J. 共1996兲. ‘‘Modeling strong discontinuities in solid mechanics via
Lemaitre, J., and Chaboche, J.-L. 共1990兲. Mechanics of solid materials, strain softening constitutive equations. Part 1: Fundamentals. Part 2:
Cambridge University Press, Cambridge, U.K. Numerical simulation.’’ Int. J. Numer. Methods Eng., 39, 3575–3624.
Lippmann, H. 共1969兲. ‘‘Eine Cosserat Theorie des plastischen Fließens,’’ Oliver, J., Cervera, M., and Manzoli, O. 共1998兲. ‘‘On the use of strain-
Acta Mech., 8, 255–284. softening models for the simulation of strong discontinuities in sol-
Lotfi, H. R., and Shing, P. B. 共1995兲. ‘‘Embedded representation of frac- ids.’’ Material instabilities in solids, R. de Borst and E. van der Gies-
ture in concrete with mixed finite elements.’’ Int. J. Numer. Methods sen, eds., Wiley, Chichester, U.K., 107–123.
Eng., 38, 1307–1325. Olofsson, T., Klisinski, M., and Nedar, P. 共1994兲. ‘‘Inner softening bands:
Luciano, R., and Willis, J. R. 共2001兲. ‘‘Nonlocal constitutive response of A new approach to localization in finite elements.’’ Computational
a random laminate subjected to configuration-dependent body force.’’ modeling of concrete structures, H. Mang, N. Bićanić, and R. de
J. Mech. Phys. Solids, 49, 431– 444. Borst, eds., Pineridge, Swansea, U.K., 373–382.
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

Ma, Q., and Clarke, D. R. 共1995兲. ‘‘Size-dependent hardness in silver Ortiz, M. 共1988兲. ‘‘Microcrack coalescence and macroscopic crack
single crystals.’’ J. Mater. Res., 10, 853– 863. growth initiation in brittle solids.’’ Int. J. Solids Struct., 24, 231–250.
Mariotte, E. 共1686兲. Traité du mouvement des eaux, M. de la Hire, ed., Ortiz, M., Leroy, Y., and Needleman, A. 共1987兲. ‘‘A finite-element
English translation by J. T. Desvaguliers, London 共1718兲. method for localized failure analysis.’’ Comput. Methods Appl. Mech.
Mazars, J. 共1984兲. ‘‘Application de la mécanique de l’endommagement Eng., 61, 189–214.
au comportement nonlinéaire et à la rupture du béton de structure.’’ Ortiz, M., and Pandolfi, A. 共1999兲. ‘‘Finite-deformation irreversible co-
Thèse de Doctorat d’Etat, Univ. Paris VI, France. hesive elements for three-dimensional crack-propagation analysis.’’
Mazars, J., and Berthaud, Y. 共1989兲. ‘‘Une technique expérimentale ap- Int. J. Numer. Methods Eng., 44, 1267–1282.
pliquée au béton pour créer un endommagement diffus et mettre en Oseen, C. W. 共1933兲. ‘‘The theory of liquid crystals.’’ Trans. Faraday
évidence son caractère unilateral.’’ Compt. Rendus Acad. Sci., Paris, Soc., 29, 883– 899.
308, 579–584. Otsuka, K., and Date, H. 共2000兲. ‘‘Fracture process zone in concrete
Mazars, J., Berthaud, Y., and Ramtani, S. 共1990兲. ‘‘The unilateral behav- tension specimen.’’ Eng. Fract. Mech., 65, 111–131.
ior of damaged concrete.’’ Eng. Fract. Mech., 35, 629– 635. Ožbolt, J., and Bažant, Z. P. 共1996兲. ‘‘Numerical smeared fracture analy-
Melenk, J. M., and Babuška, I. 共1996兲. ‘‘The partition of unity finite- sis: Nonlocal microcrack interaction approach.’’ Int. J. Numer. Meth-
element method: Basic theory and applications.’’ Comput. Methods
ods Eng., 39, 635– 661.
Appl. Mech. Eng., 39, 289–314.
Pandolfi, A., Krysl, P., and Ortiz, M. 共1999兲. ‘‘Finite-element simulation
Mihashi, H., and Nomura, N. 共1996兲. ‘‘Correlation between characteris-
of ring expansion and fragmentation: The capturing of length and time
tics of fracture process zone and tension-softening properties of con-
scales through cohesive models of fracture.’’ Int. J. Fract., 95, 279–
crete.’’ Nucl. Eng. Des., 165, 359–376.
297.
Mindlin, R. D. 共1964兲. ‘‘Microstructure in linear elasticity.’’ Arch. Ration.
Peerlings, R. H. J., de Borst, R., Brekelmans, W. A. M., and de Vree, J. H.
Mech. Anal., 16, 51–78.
P. 共1996兲. ‘‘Gradient-enhanced damage for quasibrittle materials.’’ Int.
Mindlin, R. D. 共1965兲. ‘‘Second gradient of strain and surface tension in
J. Numer. Methods Eng., 39, 3391–3403.
linear elasticity.’’ Int. J. Solids Struct., 1, 417– 438.
Pietruszczak, S., and Mróz, Z. 共1981兲. ‘‘Finite-element analysis of defor-
Mindlin, R. D., and Tiersten, H. F. 共1962兲. ‘‘Effects of couple stresses in
mation of strain-softening materials.’’ Int. J. Numer. Methods Eng.,
linear elasticity.’’ Arch. Ration. Mech. Anal., 11, 415– 448.
17, 327–334.
Moës, N., and Belytschko, T. 共2002兲. ‘‘Extended finite-element method
Pijaudier-Cabot, G., and Bažant, Z. P. 共1987兲. ‘‘Nonlocal damage theory.’’
for cohesive crack growth.’’ Eng. Fract. Mech., 69, 813– 833.
J. Eng. Mech., 113共10兲, 1512–1533.
Moës, N., Dolbow, J., and Belytschko, T. 共1999兲. ‘‘A finite-element
Planas, J., and Elices, M. 共1992兲. ‘‘Asymptotic analysis of a cohesive
method for crack growth without remeshing.’’ Int. J. Numer. Methods
crack: 1. Theoretical background.’’ Int. J. Fract., 55, 153–177.
Eng., 46, 131–150.
Planas, J., and Elices, M. 共1993兲. ‘‘Asymptotic analysis of a cohesive
Morrison, J. L. M. 共1939兲. ‘‘The yield of mild steel with particular refer-
crack: 2. Influence of the softening curve.’’ Int. J. Fract., 64, 221–
ence to the effect of size of specimen.’’ Proc. Inst. Mech. Eng., 142,
237.
193–223.
Naghdi, P. M., and Trapp, J. A. 共1975兲. ‘‘The significance of formulating Planas, J., Elices, M., and Guinea, G. V. 共1993兲. ‘‘Cohesive cracks versus
plasticity theory with reference to loading surfaces in strain space.’’ nonlocal models: Closing the gap.’’ Int. J. Fract., 63, 173–187.
Int. J. Eng. Sci., 13, 785–797. Planas, J., Guinea, G. V., and Elices, M. 共1996兲. ‘‘Basic issues on nonlo-
Needleman, A. 共1987兲. ‘‘A continuum model for void nucleation by in- cal models: Uniaxial modeling.’’ Tech. Rep. No. 96-jp03, Departa-
clusion debonding.’’ J. Appl. Mech., 54, 525–531. mento de Ciencia de Materiales, ETS de Ingenieros de Caminos,
Needleman, A. 共1988兲. ‘‘Material rate dependence and mesh sensitivity in Univ. Politécnica de Madrid, Ciudad Univ. sn., 28040 Madrid, Spain.
localization problems.’’ Comput. Methods Appl. Mech. Eng., 67, 69– Polizzotto, C. 共2001兲. ‘‘Nonlocal elasticity and related variational prin-
85. ciples.’’ Int. J. Solids Struct., 38, 7359–7380.
Nilsson, C. 共1994兲. ‘‘On nonlocal plasticity, strain softening, and local- Polizzotto, C. 共2002兲. ‘‘Remarks on some aspects of nonlocal theories in
ization.’’ Rep. No. TVSM-1007, Division of Structural Mechanics, solid mechanics.’’ Proc., 6th National Congr. SIMAI, Chia Laguna,
Lund Institute of Technology, Lund, Sweden. Italy, CD-ROM.
Nilsson, C. 共1997兲. ‘‘Nonlocal strain softening bar revisited.’’ Int. J. Sol- Polizzotto, C., Borino, G., and Fuschi, P. 共1998兲. ‘‘A thermodynamic
ids Struct., 34, 4399– 4419. consistent formulation of nonlocal and gradient plasticity.’’ Mech.
Nilsson, C. 共1999兲. ‘‘Author’s closure.’’ Int. J. Solids Struct., 36, 3093– Res. Commun., 25, 75– 82.
3100. Poole, W. J., Ashby, M. F., and Fleck, N. A. 共1996兲. ‘‘Microhardness of
Nix, W. D. 共1989兲. ‘‘Mechanical properties of thin films.’’ Metall. Trans. annealed and work-hardened copper polycrystals.’’ Scr. Mater., 34,
A, 20A, 2217–2245. 559–564.
Noll, W. 共1972兲. ‘‘A new mathematical theory of simple materials.’’ Arch. Rajagopal, E. S. 共1960兲. ‘‘The existence of interfacial couples in infini-
Ration. Mech. Anal., 48, 1–50. tesimal elasticity.’’ Ann. Phys. (Leipzig), 6, 192–201.
Nooru-Mohamed, M. B. 共1992兲. ‘‘Mixed-mode fracture of concrete: An Rashid, Y. R. 共1968兲. ‘‘Analysis of prestressed concrete pressure vessels.’’
experimental approach.’’ PhD thesis, Delft Univ. of Technology, Delft, Nucl. Eng. Des., 7, 334 –344.
The Netherlands. Rayleigh, O. M. 共1918兲. ‘‘Notes on the theory of lubrication.’’ Philos.
Ohlsson, U., and Olofsson, T. 共1997兲. ‘‘Mixed-mode fracture and anchor Mag., 35, 1–12.
bolts in concrete: Analysis with inner softening bands.’’ J. Eng. Read, H. E., and Hegemier, G. A. 共1984兲. ‘‘Strain softening of rock, soil,
Mech., 123共10兲, 1027–1033. and concrete—A review article.’’ Mech. Mater., 3, 271–294.

1148 / JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002

J. Eng. Mech., 2002, 128(11): 1119-1149

 Richards, C. W. 共1958兲. ‘‘Effects of size on the yielding of mild steel Svedberg, T. 共1996兲. ‘‘A thermodynamically consistent theory of
beams.’’ Proc. Am. Soc. Test. Mater., 58, 955–970. gradient-regularized plasticity coupled to damage.’’ Licentiate thesis,
Rodrı́guez-Ferran, A., and Huerta, A. 共2000兲. ‘‘Error estimation and adap- Chalmers Univ. of Technology.
tivity for nonlocal damage models.’’ Int. J. Solids Struct., 37, 7501– Svedberg, T., and Runesson, K. 共1998兲. ‘‘Thermodynamically consistent
7528. nonlocal and gradient formulations of plasticity.’’ Nonlocal aspects in
Rogula, D. 共1965兲. ‘‘Influence of spatial acoustic dispersion on dynamical solid mechanics, A. Brillard and J. F. Ganghoffer, eds., EUROMECH
properties of dislocations. I.’’ Bull. Acad. Pol. Sci., Ser. Sci. Tech., 13, Colloquium 378, Mulhouse, France, 32–37.
337–343. Thomas, T. Y. 共1961兲. Plastic flow and fracture of solids, Academic, New
Rogula, D. 共1982兲. ‘‘Introduction to nonlocal theory of material media.’’ York.
Nonlocal theory of material media, CISM courses and lectures, D. Toupin, R. A. 共1962兲. ‘‘Elastic materials with couple stresses.’’ Arch.
Rogula, ed., Springer, Wien, 268, 125–222. Ration. Mech. Anal., 11, 385– 414.
Downloaded from ascelibrary.org by "Indian Institute of Technology, Gandhinagar" on 01/24/24. Copyright ASCE. For personal use only; all rights reserved.

Rolshoven, S., and Jirásek, M. 共2001兲. ‘‘On regularized plasticity models Toupin, R. A. 共1964兲. ‘‘Theories of elasticity with couple stress.’’ Arch.
for strain-softening materials.’’ Fracture mechanics of concrete struc- Ration. Mech. Anal., 17, 85–112.
tures, R. de Borst, J. Mazars, G. Pijaudier-Cabot, and J. G. M. van Truesdell, C., and Toupin, R. A. 共1960兲. ‘‘Classical field theories of me-
Mier, eds., Balkema, Lisse, 617– 624. chanics.’’ Handbuch der physik, Springer, Berlin, III, 1.
Rossi, P., and Wu, X. 共1992兲. ‘‘Probabilistic model for material behavior Tvergaard, V., and Hutchinson, J. W. 共1992兲. ‘‘The relation between crack
analysis and appraisement of concrete structures.’’ Mag. Concrete growth resistance and fracture process parameters in elastic–plastic
Res., 44, 271–280. solids.’’ J. Mech. Phys. Solids, 40, 1377–1397.
Rots, J. G. 共1988兲. ‘‘Computational modeling of concrete fracture.’’ PhD Tvergaard, V., and Hutchinson, J. W. 共1993兲. ‘‘The influence of plasticity
thesis, Delft Univ. of Technol., Delft, The Netherlands. on mixed mode interface toughness.’’ J. Mech. Phys. Solids, 41,
Sandler, I. S. 共1984兲. ‘‘Strain-softening for static and dynamic problems.’’ 1119–1135.
Proc., Symp. on Constitutive Equations, ASME, Winter Annual Meet- Valanis, K. C. 共1990兲. ‘‘A theory of damage in brittle materials.’’ Eng.
ing, New Orleans, 217–231.
Fract. Mech., 36, 403– 416.
Saouridis, C. 共1988兲. ‘‘Identification et numérisation objectives des com-
Valanis, K. C. 共1991兲. ‘‘A global damage theory and the hyperbolicity of
portements adoucissants: Une approche multiéchelle de
the wave problem.’’ J. Appl. Mech., 58, 311–316.
l’endommagement du béton.’’ PhD thesis, Univ. Paris VI.
van Gils, M. A. J. 共1997兲. ‘‘Quasibrittle fracture of ceramics.’’ PhD thesis,
Saouridis, C., and Mazars, J. 共1992兲. ‘‘Prediction of the failure and size
Eindhoven Univ. of Technology, The Netherlands.
effect in concrete via a biscale damage approach.’’ Eng. Comput., 9,
van Mier, J. G. M. 共1997兲. Fracture processes of concrete, CRC, Boca
329–344.
Raton, Fla.
Schlangen, E. 共1993兲. ‘‘Experimental and numerical analysis of fracture
Vermeer, P. A., and Brinkgreve, R. B. J. 共1994兲. ‘‘A new effective non-
processes in concrete.’’ PhD thesis, Delft Univ. of Technology, Delft,
The Netherlands. local strain measure for softening plasticity.’’ Localization and bifur-
Schlangen, E., and van Mier, J. G. M. 共1992兲. ‘‘Simple lattice model for cation theory for soils and rocks, R. Chambon, J. Desrues, and I.
numerical simulation of fracture of concrete materials and structures.’’ Vardoulakis, eds., Balkema, Rotterdam, The Netherlands, 89–100.
Mater. Struct., 25, 534 –542. Voigt, W. 共1887兲. ‘‘Theoretische Studien über die Elastizitätsverhältnisse
Shi, G. H. 共1991兲. ‘‘Manifold method of material analysis.’’ Trans., 9th der Krystalle.’’ Abh. Königlichen Gesellschaft Wiss. Göttingen.
Army Conf. on Applied Mathematics and Computing, Minneapolis, Voigt, W. 共1894兲. ‘‘Über Medien ohne innere Kräfte und eine durch sie
57–76. gelieferte mechanische Deutung der Maxwell-Hertzschen Gleichun-
Simo, J. C., and Oliver, J. 共1994兲. ‘‘A new approach to the analysis and gen.’’ Abh. Königlichen Gesellschaft Wiss. Göttingen, 40, 72–79.
simulation of strain softening in solids.’’ Fracture and damage in Walsh, P. F. 共1972兲. ‘‘Fracture of plain concrete.’’ Indian Concr. J., 46,
quasibrittle structures, Z. P. Bažant, Z. Bittnar, M. Jirásek, and J. 469– 470 and 476.
Mazars, eds., E&FN Spon, London, 25–39. Walsh, P. F. 共1976兲. ‘‘Crack initiation in plain concrete.’’ Mag. Concrete
Simo, J. C., Oliver, J., and Armero, F. 共1993兲. ‘‘An analysis of strong Res., 28, 37– 41.
discontinuities induced by strain softening in rate-independent inelas- Wei, Y., and Hutchinson, J. W. 共1999兲. ‘‘Models of interface separation
tic solids.’’ Computational Mech., Berlin, 12, 277–296. accompanied by plastic dissipation at multiple scales.’’ Int. J. Fract.,
Sluys, L. J. 共1997兲. ‘‘Discontinuous modeling of shear banding.’’ Com- 95, 1–17.
putational plasticity: Fundamentals and applications, D. R. J. Owen, Weibull, W. 共1939兲. ‘‘The phenomenon of rupture in solids.’’ Proc. R.
E. Oñate, and E. Hinton, eds., Int. Center for Numerical Methods in Swedish Inst. Eng. Res. (Ing. Akad. Handl.), 153, 1–55.
Eng., Barcelona, Spain, 735–744. Wells, G. N. 共2001兲. ‘‘Discontinuous modeling of strain localization and
Stolken, J. S., and Evans, A. G. 共1998兲. ‘‘A microbend test method for failure.’’ PhD thesis, Delft Univ. of Technol., Delft, The Netherlands.
measuring the plasticity length scale.’’ Acta Mater., 46, 5109–5115. Willam, K., Bićanić, N., and Sture, S. 共1986兲. ‘‘Composite fracture model
Strömberg, L., and Ristinmaa, M. 共1996兲. ‘‘FE formulation of a nonlocal for strain-softening and localized failure of concrete.’’ Computational
plasticity theory.’’ Comput. Methods Appl. Mech. Eng., 136, 127–144. modeling of reinforced concrete structures, E. Hinton and D. R. J.
Sukumar, N., Moës, N., Moran, B., and Belytschko, T. 共2000兲. ‘‘Extended Owen, eds., Pineridge, Swansea, 122–153.
finite-element method for three-dimensional crack modeling.’’ Int. J. Xu, X.-P., and Needleman, A. 共1994兲. ‘‘Numerical simulations of fast
Numer. Methods Eng., 48, 1549–1570. crack growth in brittle solids.’’ J. Mech. Phys. Solids, 42, 1397–1434.

JOURNAL OF ENGINEERING MECHANICS / NOVEMBER 2002 / 1149

J. Eng. Mech., 2002, 128(11): 1119-1149